影像辨識 - HackMD

# 影像辨識 [TOC] <style> .flex-container{ display:flex; justify-content:center; } </style> ## 壹、安裝環境 ### (一)軟體下載安裝 #### 1. 至[這裡](https://github.com/ultralytics/yolov5)，下載yolov5原始程式碼。 :::warning 需要配合安裝Python 3.6以上版本 ::: <div class="flex-container"> <img src="https://i.imgur.com/V69D7Wo.jpg" width="100%"></div> #### 2. 至[Anaconda官網](https://www.anaconda.com/products/individual)下載。 <div class="flex-container"> <img src="https://i.imgur.com/nTtguQ0.jpg" width="90%"></div> :::warning ！！兩個選項都要打勾！！ ::: <div class="flex-container"> <img src="https://i.imgur.com/DRBo9sk.jpg" width="70%"></div> #### 3. 至[PyCharm下載](https://www.jetbrains.com/pycharm/download/#section=windows)Community。 <div class="flex-container"> <img src="https://i.imgur.com/rke0jeY.jpg" width="100%"></div> --- ### (二)新增虛擬環境 #### 1. anaconda安裝完後進入environment介面，點選create。 <div class="flex-container"> <img src="https://i.imgur.com/in2bFQy.jpg" width="100%"></div> #### 2. 建立新環境並命名，在此命名為「yolov5」；Python版本（先前下載的版本）選3.8後，點選create。 <div class="flex-container"> <img src="https://i.imgur.com/pDLRID5.jpg" width="90%"></div> #### 3. 第二步驟建立成功後environment介面會出現剛才建立好的「yolov5」。 <div class="flex-container"> <img src="https://i.imgur.com/N0LO40N.jpg" width="100%"></div> #### 4. 回到home介面，點選PyCharm Community。 :::warning ！！注意位置要在yolov5裡！！ ::: <div class="flex-container"> <img src="https://i.imgur.com/IuEiEDL.jpg" width="100%"></div> #### 5. 進入PyCharm，點選open。 <div class="flex-container"> <img src="https://i.imgur.com/1HGQRPN.jpg" width="100%"></div> #### 6. 選擇yolov5-master資料夾。 <div class="flex-container"> <img src="https://i.imgur.com/FKYSvMw.jpg" width="70%"></div> #### 7. 進入PyCharm後，打開File。 <div class="flex-container"> <img src="https://i.imgur.com/JAIJO0Z.jpg" width="100%"></div> #### 8. 點選File裡的「Settings...」。 <div class="flex-container"> <img src="https://i.imgur.com/DDk77Kf.jpg" width="50%"></div> #### 9. 找到「Project」後，點選「Python Interpreter」，按下設定圖示。 <div class="flex-container"> <img src="https://i.imgur.com/wYUwAOt.jpg" width="100%"></div> #### 10. 選擇「Add...」。 <div class="flex-container"> <img src="https://i.imgur.com/cWyR8ou.jpg" width="100%"></div> #### 11. 在「Conda Environment」裡點選「Existing environment」，確認路徑是剛才建立的虛擬環境「yolov5」後按下「OK」。 <div class="flex-container"> <img src="https://i.imgur.com/w09hh74.jpg" width="100%"></div> --- ### (三)安裝常用工具 #### 1. 在anaconda的home介面裡，Install「CMD.exe Prompt」。 <div class="flex-container"> <img src="https://i.imgur.com/W6q8xNt.jpg" width="100%"></div> #### 2. 點選「CMD.exe Prompt」。 <div class="flex-container"> <img src="https://i.imgur.com/UPn7Xbb.jpg" width="100%"></div> #### 3. 進入「CMD.exe Prompt」後，進行指令指令輸入。 ``` conda install git ``` <div class="flex-container"> <img src="https://i.imgur.com/RRT0DMQ.jpg" width="90%"></div> #### 4.上步驟指令跑完後，再執行下方指令。 ``` conda install pip ``` <div class="flex-container"> <img src="https://i.imgur.com/03Xsdvz.jpg" width="90%"></div> :::info 第3、4步驟都需時間執行。 ::: #### 5. 回到Environments，點選「yolov5」，並更新擴充軟體。 <div class="flex-container"> <img src="https://i.imgur.com/at7DirS.jpg" width="100%"></div> #### 6. 下載opencv。 <div class="flex-container"> <img src="https://i.imgur.com/aXHXz3P.jpg" width="100%"></div> --- ### (四)安裝Cuda & Cudnn #### 1. 尋找自己電腦GPU。 <div class="flex-container"> <img src="https://i.imgur.com/3x3hY6M.jpg" width="80%"></div> #### 2. 更新[NVDIA DRIVER](https://www.nvidia.com.tw/Download/index.aspx?lang=tw)。 :::warning 記得「產品家族」要選擇跟自己電腦GPU一樣的 ::: <div class="flex-container"> <img src="https://i.imgur.com/St7nIZf.jpg" width="90%"></div> #### 3. 點[這裡](https://en.wikipedia.org/wiki/CUDA)參考Cuda GPU支援狀況。 :::warning GPU支援狀況，數字越大，運算越快，原則上3以上就沒問題了 ::: <div class="flex-container"> <img src="https://i.imgur.com/wjFH9xo.jpg" width="100%"></div> #### 4. 下載[Cuda10.2](https://developer.nvidia.com/cuda-10.2-download-archive?target\_os=Windows&target\_arch=x86\_64&target\_version=10&target_type=exelocal)。 <div class="flex-container"> <img src="https://i.imgur.com/BVaRRI8.jpg" width="100%"></div> #### 5. 建在C槽下。 <div class="flex-container"> <img src="https://i.imgur.com/d7tl9kM.jpg" width="80%"></div> #### 6. 下載[Cudnn10.2](https://developer.nvidia.com/rdp/cudnn-download)。 <div class="flex-container"> <img src="https://i.imgur.com/wu9DJ24.jpg" width="100%"></div> #### 7. 解壓縮到C槽。 <div class="flex-container"> <img src="https://i.imgur.com/TbqqIpz.jpg" width="90%"></div> #### 8. 點開「cuda」資料夾後可以看到如下圖內容。 <div class="flex-container"> <img src="https://i.imgur.com/SeEEzaz.jpg" width="90%"></div> #### 9. 找到如以下圖片的路徑。 <div class="flex-container"> <img src="https://i.imgur.com/LjvZ9uU.jpg" width="100%"></div> #### 10. 將8的bin裡面的內容複製到9的bin。 <div class="flex-container"> <img src="https://i.imgur.com/CCh7Y11.jpg" width="100%"></div> #### 11. 將8的include裡面的內容複製到9的include。 <div class="flex-container"> <img src="https://i.imgur.com/rSayQl7.jpg" width="100%"></div> #### 12. 將8的\lib\x64裡面的內容複製到9的\lib\x64。 <div class="flex-container"> <img src="https://i.imgur.com/E7A3Mjk.jpg" width="100%"></div> #### 13.新增環境變數。 <div class="flex-container"> <img src="https://i.imgur.com/5l3vLMt.jpg" width="90%"></div> #### 14. 點選「新增」後，如下輸入。 ``` C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\lib\x64 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\include C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\extras\CUPTI\lib64 C:\ProgramData\NVIDIA Corporation\CUDA Samples\v10.2\bin\win64 C:\ProgramData\NVIDIA Corporation\CUDA Samples\v10.2\common\lib\x64 ``` <div class="flex-container"> <img src="https://i.imgur.com/B6nGNbC.jpg" width="100%"></div> #### 15.確定系統變數有如下圖的兩個。 <div class="flex-container"> <img src="https://i.imgur.com/jYOdv8N.jpg" width="100%"></div> #### 16. 打開cmd，指令輸入進行驗證。 ``` nvcc -V ``` <div class="flex-container"> <img src="https://i.imgur.com/s4dHkYr.jpg" width="100%"></div> --- ### (五)安裝PyTorch :::warning PyTorch版本必須選擇1.7.0以上 ::: #### 1. 相關[PyTorch](https://pytorch.org/get-started/previous-versions/)指令，照著以下步驟輸入指令。 ``` activate yolov5 ``` <div class="flex-container"> <img src="https://i.imgur.com/YA9MW7w.jpg" width="90%"></div> #### 2. 選擇v1.7.1版本裡，複製「CUDA 10.2」安裝指令。 ``` # CUDA 10.2 conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch ``` <div class="flex-container"> <img src="https://i.imgur.com/OxZIbjA.jpg" width="100%"></div> :::info 此步驟需時間執行。 ::: #### 3. 打開cmd，指令輸入進行驗證。(可依照下面程式碼複製) ``` activate yolov5 python import torch as t t.__version__ ``` <div class="flex-container"> <img src="https://i.imgur.com/AXBx4to.jpg" width="100%"></div> --- ### (六)安裝requirements文件內容 #### 1. 到「yolov5_master」的資料夾裡找尋「requirements」的文件，並複製紅框的內容。 <div class="flex-container"> <img src="https://i.imgur.com/NxlZOMr.jpg" width="100%"></div> #### 2. 將複製的複製的內容貼在PyCharm的terminal裡。 <div class="flex-container"> <img src="https://i.imgur.com/2BQ6ddR.jpg" width="100%"></div> --- ### (七)下載權重檔案 #### 1. 至[這裡](https://github.com/ultralytics/yolov5/releases/tag/v5.0)下載權重檔案 <div class="flex-container"> <img src="https://i.imgur.com/UIpC1Yi.jpg" width="50%"></div> #### 2. 放在「yolov5-master」資料夾下 <div class="flex-container"> <img src="https://i.imgur.com/MPz4y6A.jpg" width="100%"></div> --- ### (八)測試 #### 1. 執行`detect.py` <div class="flex-container"> <img src="https://i.imgur.com/u4FmWIf.jpg" width="100%"></div> #### 2. 測試照片影片存放路徑。 <div class="flex-container"> <img src="https://i.imgur.com/QUFkitn.jpg" width="100%"></div> #### 3. 測試"結果"存放路徑。 <div class="flex-container"> <img src="https://i.imgur.com/F57bjTn.jpg" width="100%"></div> #### 4. 測試結果 a. <div class="flex-container"> <img src="https://i.imgur.com/rpgmQLj.jpg" width="80%"></div> b. <div class="flex-container"> <img src="https://i.imgur.com/hbkKJto.jpg" width="70%"></div> c. <div class="flex-container"> <img src="https://i.imgur.com/h8fjlnB.jpg" width="70%"></div> d. <div class="flex-container"> <img src="https://i.imgur.com/u4Fg37M.jpg" width="70%"></div> e. <div class="flex-container"> <img src="https://i.imgur.com/5JqzAkY.jpg" width="80%"></div> 5.連接webcam測試 <div class="flex-container"> <img src="https://i.imgur.com/06YC8AI.jpg" width="100%"></div> --- ## 貳、訓練 ### (一)安裝pycocotools #### 1. 點此下載[Microsoft Visual C++ Build Tools 2015](https://www.microsoft.com/zh-TW/download/details.aspx?id=48159)。 <div class="flex-container"> <img src="https://i.imgur.com/Pu7gBC3.jpg" width="90%"></div> #### 2. 點此[下載pycocotools的.whl檔案](https://pypi.tuna.tsinghua.edu.cn/simple/pycocotools-windows/)。 (原始檔：https://github.com/cocodataset/cocoapi) <div class="flex-container"> <img src="https://i.imgur.com/Y5pjswm.jpg" width="70%"></div> #### 3. 存放路徑。 <div class="flex-container"> <img src="https://i.imgur.com/3xRFeN1.jpg" width="100%"></div> #### 4. 打開cmd後，進行指令輸入。 ``` activate yolov5 pip install C:\Users\Ching\anaconda3\lib\pycocotools_windows-2.0.0.2-cp38-cp38-win_amd64.whl ``` :::warning pip install 後的路徑為pycocotools_windows-2.0.0.2-cp38-cp38-win_amd64.whl的存放路徑。 ::: #### 5. ``` pip install scikit-image ``` <div class="flex-container"> <img src="https://i.imgur.com/UHNNQJb.jpg" width="100%"></div> #### 6. ``` pip intall pycocotools ``` <div class="flex-container"> <img src="https://i.imgur.com/jFc5jbz.jpg" width="100%"></div> #### 7. 指令輸入進行驗證。輸入```pip list```，確認pycocotools、scikit-image。 <div class="flex-container"> <img src="https://i.imgur.com/AkzoYbU.jpg" width="70%"></div> ### (二)安裝apex #### 1. 點此[下載apex原檔](https://github.com/NVIDIA/apex)。 <div class="flex-container"> <img src="https://i.imgur.com/ZxxB3uR.jpg" width="100%"></div> #### 2. 存放位置。 <div class="flex-container"> <img src="https://i.imgur.com/dyDonq7.jpg" width="80%"></div> #### 3. 打開cmd，進行指令輸入。 ``` activate yolov5 cd apex-master python setup.py install ``` <div class="flex-container"> <img src="https://i.imgur.com/JwMvcVl.jpg" width="100%"></div> #### 4. 指令輸入進行驗證。輸入```pip list```，有顯示apex表安裝成功。 <div class="flex-container"> <img src="https://i.imgur.com/rgbPCW8.jpg" width="100%"></div> ### (三)標記資料集 #### 1. 新增資料夾。 <div class="flex-container"> <img src="https://i.imgur.com/nNGGrXe.jpg" width="100%"></div> :::info 說明： [data] Annotations：存放的是標記後生成的xml檔案，進行 detection 任務時的標籤檔案，xml 形式，檔名與圖片名一一對應。 images：存放 .jpg 格式的圖片檔案。 ImageSets：存放的是訓練資料集和測試資料集的分類情況，是分類和檢測的資料集分割檔案，包含train.txt, val.txt,trainval.txt,test.txt。 labels：存放的是儲存標記內容的txt檔案，label標註資訊的txt檔案，與圖片一一對應。 ----- [ImageSets] (train，val，test建議按照8：1：1比例劃分) train.txt：寫著用於訓練的圖片名稱。 val.txt：寫著用於驗證的圖片名稱。 trainval.txt：train與val的合集。 test.txt：寫著用於測試的圖片名稱。 ::: #### 2.點此下載[精靈標記助手(Colabeler)](http://www.jinglingbiaozhu.com/)。 <div class="flex-container"> <img src="https://i.imgur.com/vkjI2Qn.jpg" width="100%"></div> #### 3.照片存放路徑（D:/yolov5-master/data/images）。 <div class="flex-container"> <img src="https://i.imgur.com/RPJt1sA.jpg" width="100%"></div> #### 4.新建->位置標注->分類值(辨識人臉名字)->照片存放路徑，確認無誤按「創建」。 <div class="flex-container"> <img src="https://i.imgur.com/hCElt69.jpg" width="100%"></div> #### 5.框臉後在右邊標注信息欄選擇相對應的名字，並按下下方「打勾」進行儲存，每框完一張都要按完打勾才能跳下一張，不然剛才框的資料會不見。 <div class="flex-container"> <img src="https://i.imgur.com/O4yBSeu.jpg" width="100%"></div> #### 6.全部照片都框完以後，按下左下方「導出」，輸出方式選擇「pascal-voc」。 <div class="flex-container"> <img src="https://i.imgur.com/PjgqMwB.jpg" width="100%"></div> #### 7.導出後的xml檔案存放路徑。 <div class="flex-container"> <img src="https://i.imgur.com/fnZadpW.jpg" width="100%"></div> ### (四)構建資料集 #### 1. 在yolov5-master的根目錄下新建檔案 `makeTxt.py`、`voc_label.py` 建立完成後執行這兩個檔案 :::info 「makeTxt.py」 makeTxt.py主要是將資料集分類成訓練資料集和測試資料集。預設train，val，test按照8：1：1的比例進行隨機分類。執行後ImagesSets資料夾中會出現四個檔案，主要是生成的訓練資料集和測試資料集的圖片名稱，同時data目錄下也會出現這四個檔案，內容是訓練資料集和測試資料集的圖片路徑。 ::: :::info 「voc_label.py」 classes=[……] ：填入自己在資料集中所標註的類別名稱，填寫錯誤的話會造成讀取不出xml檔案裡的標註資訊。 voc_label.py主要是將圖片資料集標註後的xml檔案中的標註資訊讀取出來並寫入txt檔案。執行後在labels資料夾中出現所有圖片資料集的標註資訊。 ::: #### 2. makeTxt執行後ImagesSets資料夾中會出現四個檔案。 <div class="flex-container"> <img src="https://i.imgur.com/G85mIPf.jpg" width="100%"></div> #### 3. voc_label執行後labels資料夾中出現所有圖片的標註資訊。 <div class="flex-container"> <img src="https://i.imgur.com/iWWwzOi.jpg" width="100%"></div> #### 4. makeTxt程式檔。 ```python= import os import random trainval_percent = 0.9 train_percent = 0.9 xmlfilepath = 'data/Annotations' txtsavepath = 'data/ImageSets' total_xml = os.listdir(xmlfilepath) num = len(total_xml) list = range(num) tv = int(num * trainval_percent) tr = int(tv * train_percent) trainval = random.sample(list, tv) train = random.sample(trainval, tr) ftrainval = open('data/ImageSets/trainval.txt', 'w') ftest = open('data/ImageSets/test.txt', 'w') ftrain = open('data/ImageSets/train.txt', 'w') fval = open('data/ImageSets/val.txt', 'w') for i in list: name = total_xml[i][:-4] + '\n' if i in trainval: ftrainval.write(name) if i in train: ftrain.write(name) else: fval.write(name) else: ftest.write(name) ftrainval.close() ftrain.close() fval.close() ftest.close() ``` #### 5. voc_label程式檔。 ```python= # xml解析包 import xml.etree.ElementTree as ET import pickle import os # os.listdir() 方法用於返回指定的資料夾包含的檔案或資料夾的名字的列表 from os import listdir, getcwd from os.path import join sets = ['train', 'test', 'val'] classes = ['Angel', 'Sophia', 'Jenny', 'Jim', 'Tina'] # 進行歸一化操作 def convert(size, box): # size:(原圖w,原圖h) , box:(xmin,xmax,ymin,ymax) dw = 1./size[0] # 1/w dh = 1./size[1] # 1/h x = (box[0] + box[1])/2.0 # 物體在圖中的中心點x座標 y = (box[2] + box[3])/2.0 # 物體在圖中的中心點y座標 w = box[1] - box[0] # 物體實際畫素寬度 h = box[3] - box[2] # 物體實際畫素高度 x = x*dw # 物體中心點x的座標比(相當於 x/原圖w) w = w*dw # 物體寬度的寬度比(相當於 w/原圖w) y = y*dh # 物體中心點y的座標比(相當於 y/原圖h) h = h*dh # 物體寬度的寬度比(相當於 h/原圖h) return (x, y, w, h) # 返回相對於原圖的物體中心點的x座標比,y座標比,寬度比,高度比,取值範圍[0-1] # year ='2012', 對應圖片的id（檔名） def convert_annotation(image_id): ''' 將對應檔名的xml檔案轉化為label檔案，xml檔案包含了對應的bunding框以及圖片長款大小等資訊，通過對其解析，然後進行歸一化最終讀到label檔案中去，也就是說一張圖片檔案對應一個xml檔案，然後通過解析和歸一化，能夠將對應的資訊儲存到唯一一個label檔案中去 labal檔案中的格式：calss x y w h　　同時，一張圖片對應的類別有多個，所以對應的ｂｕｎｄｉｎｇ的資訊也有多個 ''' # 對應的通過year 找到相應的資料夾，並且開啟相應image_id的xml檔案，其對應bund檔案 in_file = open('data/Annotations/%s.xml' % (image_id), encoding='utf-8') # 準備在對應的image_id 中寫入對應的label，分別為 # <object-class> <x> <y> <width> <height> out_file = open('data/labels/%s.txt' % (image_id), 'w', encoding='utf-8') # 解析xml檔案 tree = ET.parse(in_file) # 獲得對應的鍵值對 root = tree.getroot() # 獲得圖片的尺寸大小 size = root.find('size') # 如果xml內的標記為空，增加判斷條件 if size != None: # 獲得寬 w = int(size.find('width').text) # 獲得高 h = int(size.find('height').text) # 遍歷目標obj for obj in root.iter('object'): # 獲得difficult ？？ difficult = obj.find('difficult').text # 獲得類別 =string 型別 cls = obj.find('name').text # 如果類別不是對應在我們預定好的class檔案中，或difficult==1則跳過 if cls not in classes or int(difficult) == 1: continue # 通過類別名稱找到id cls_id = classes.index(cls) # 找到bndbox 物件 xmlbox = obj.find('bndbox') # 獲取對應的bndbox的陣列 = ['xmin','xmax','ymin','ymax'] b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text), float(xmlbox.find('ymax').text)) print(image_id, cls, b) # 帶入進行歸一化操作 # w = 寬, h = 高， b= bndbox的陣列 = ['xmin','xmax','ymin','ymax'] bb = convert((w, h), b) # bb 對應的是歸一化後的(x,y,w,h) # 生成 calss x y w h 在label檔案中 out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n') # 返回當前工作目錄 wd = getcwd() print(wd) for image_set in sets: ''' 對所有的檔案資料集進行遍歷做了兩個工作：　　　　１．將所有圖片檔案都遍歷一遍，並且將其所有的全路徑都寫在對應的txt檔案中去，方便定位　　　　２．同時對所有的圖片檔案進行解析和轉化，將其對應的bundingbox 以及類別的資訊全部解析寫到label 檔案中去　　　　　最後再通過直接讀取檔案，就能找到對應的label 資訊 ''' # 先找labels資料夾如果不存在則建立 if not os.path.exists('data/labels/'): os.makedirs('data/labels/') # 讀取在ImageSets/Main 中的train、test..等檔案的內容 # 包含對應的檔名稱 image_ids = open('data/ImageSets/%s.txt' % (image_set)).read().strip().split() # 開啟對應的2012_train.txt 檔案對其進行寫入準備 list_file = open('data/%s.txt' % (image_set), 'w') # 將對應的檔案_id以及全路徑寫進去並換行 for image_id in image_ids: list_file.write('data/images/%s.jpg\n' % (image_id)) # 呼叫 year = 年份 image_id = 對應的檔名_id convert_annotation(image_id) # 關閉檔案 list_file.close() # os.system(‘comand’) 會執行括號中的命令，如果命令成功執行，這條語句返回0，否則返回1 # os.system("cat 2007_train.txt 2007_val.txt 2012_train.txt 2012_val.txt > train.txt") # os.system("cat 2007_train.txt 2007_val.txt 2007_test.txt 2012_train.txt 2012_val.txt > train.all.txt")A ``` ### (五)檔案修改 #### 1. 首先在data目錄下，複製貼上一份coco.yaml檔案，此次訓練事先做我們組員的臉部辨識的部分，所以將其重新命名face.yaml。 ::: info 其中train，val，test後面分別為訓練集和測試集圖片的路徑，nc為資料集的類別數，我們這裡分了5類，names為類別的名稱。 ::: <div class="flex-container"> <img src="https://i.imgur.com/hg4VUEm.jpg" width="100%"></div> #### 2. face.yaml程式碼修改 <div class="flex-container"> <img src="https://i.imgur.com/0mXQyYy.jpg" width="100%"></div> #### 3. 修改yolov5s.yaml檔案。 <div class="flex-container"> <img src="https://i.imgur.com/bun2jxT.jpg" width="100%"></div> :::info 接著在models目錄下的yolov5s.yaml檔案進行修改，這裡取決於你使用了哪個模型就去修改對於的檔案，這裡使用的是yolov5s模型 ::: #### 4. 修改train.py檔案。 <div class="flex-container"> <img src="https://i.imgur.com/sDNbIdX.jpg" width="100%"></div> ### (六)開始訓練 #### 1.開始訓練。 :::info 直接執行train.py檔案開始訓練，出現此畫面就是成功開始訓練了 ::: <div class="flex-container"> <img src="https://i.imgur.com/c2xvuje.jpg" width="100%"></div> #### 2.訓練完成。 <div class="flex-container"> <img src="https://i.imgur.com/AQSAd2V.jpg" width="100%"></div> #### 3. :::info 訓練好後會在runs/train/exp資料夾得到以下檔案。我們訓練好的權重為weights資料夾中的best.pt和last.pt檔案， best.pt是訓練300輪後所得到的最好的權重，last.pt是最後一輪訓練所得到的權重。 ::: `runs\train\exp` <div class="flex-container"> <img src="https://i.imgur.com/Nl7gBf4.jpg" width="100%"></div> `runs\train\exp\weight` <div class="flex-container"> <img src="https://i.imgur.com/wfKIInY.jpg" width="100%"></div> ### (七)問題解決方法 #### 1. 分頁檔太小，無法完成操作。 :::info 將workers的默認值改為1。 ::: <div class="flex-container"> <img src="https://i.imgur.com/aRHw3mr.jpg" width="100%"></div> #### 2. ImportError DLL load failed while importing _ctypes 發生內部錯誤。 :::info 解安裝pillow後，再重新安裝一次。 ::: <div class="flex-container"> <img src="https://i.imgur.com/8dF4hx6.jpg" width="90%"></div> #### 3. RuntimeError CUDA out of memory。 :::info 將batch-size的默認值調小至電腦跑得動。 ::: <div class="flex-container"> <img src="https://i.imgur.com/CA2Me7N.jpg" width="100%"></div> ### (八)訓練結果 #### 1.在"D:/yolov5-master/data" 裡新增「imagestest」資料夾。 <div class="flex-container"> <img src="https://i.imgur.com/jC1VjmW.jpg" width="100%"></div> #### 2.複製best.pt到yolov5-master根目錄下。 <div class="flex-container"> <img src="https://i.imgur.com/P3YsU3f.jpg" width="100%"></div> #### 3.修改 detect . py。 <div class="flex-container"> <img src="https://i.imgur.com/4rSZDDw.jpg" width="100%"></div> #### 4.測試照片影片存放路徑(在新建的imagestest裡)。 <div class="flex-container"> <img src="https://i.imgur.com/BgwSGOd.jpg" width="100%"></div> #### 5.測試"結果"存放路徑。 <div class="flex-container"> <img src="https://i.imgur.com/mfgzgnu.jpg" width="100%"></div> #### 6.測試結果 <div class="flex-container"> <img src="https://i.imgur.com/vclDse9.jpg" width="70%"></div> <div class="flex-container"> <img src="https://i.imgur.com/kbGisu4.jpg" width="70%"></div> <div class="flex-container"> <img src="https://i.imgur.com/CIfSVIV.jpg" width="70%"></div> <div class="flex-container"> <img src="https://i.imgur.com/WgoXlLj.jpg" width="70%"></div> <div class="flex-container"> <img src="https://i.imgur.com/88BE6AI.jpg" width="70%"></div> <div class="flex-container"> <img src="https://i.imgur.com/MRkjuDx.jpg" width="60%"></div> :::info 下圖判斷Jenny的臉不成功 ::: <div class="flex-container"> <img src="https://i.imgur.com/C2VRotg.jpg" width="70%"></div> ### (九)tensorboard視覺化訓練 #### 1.開始訓練以後可以在"D:/yolov5-master/run/train/exp"找到以下檔案。 <div class="flex-container"> <img src="https://i.imgur.com/Bsz5A0K.jpg" width="100%"></div> #### 2.在"D:/yolov5-master"下直接輸入cmd，進入cmd。 <div class="flex-container"> <img src="https://i.imgur.com/oJyuZvc.jpg" width="100%"></div> #### 3. 指令輸入。 ``` activate yolov5 tensorboard --logdir runs/train/exp ``` <div class="flex-container"> <img src="https://i.imgur.com/AdxwRlH.jpg" width="100%"></div> #### 4. 將上一步驟得到的網址在瀏覽器開啟。 ``` http://localhost:6006/ ``` <div class="flex-container"> <img src="https://i.imgur.com/81ywKda.jpg" width="100%"></div> #### 5. 結果。 ##### a.metrics <div class="flex-container"> <img src="https://i.imgur.com/IOXKkAX.jpg" width="100%"></div> ##### b.train <div class="flex-container"> <img src="https://i.imgur.com/wuluofc.jpg" width="100%"></div> ##### c.val <div class="flex-container"> <img src="https://i.imgur.com/Ei3Ex1x.jpg" width="100%"></div> ##### d.x <div class="flex-container"> <img src="https://i.imgur.com/QxGcBqW.jpg" width="100%"></div> ### (十)權重 #### 1. 權重圖。 <div class="flex-container"> <img src="https://i.imgur.com/1gbAcbw.jpg" width="100%"></div> #### 2. YoloV5權重。 depth_multiple 代表網路的深度 width_multiple 代表網路的寬度網路的深度：yolov5s < yolov5m < yolov5l < yolov5x 網路的寬度：yolov5s < yolov5m < yolov5l < yolov5x 但在此電腦的顯示卡效能最多只能跑yolov5s、yolov5m ==yolov5l== <div class="flex-container"><img src="https://i.imgur.com/GNFvnCS.jpg" width="100%"> </div> ==yolov5m== <div class="flex-container"><img src="https://i.imgur.com/YYUhCSc.jpg" width="100%"></div> ==yolov5s== <div class="flex-container"><img src="https://i.imgur.com/AQJMNGS.jpg" width="100%"></div> ==yolov5x== <div class="flex-container"><img src="https://i.imgur.com/7va0V8S.jpg" width="100%"></div> ==統整Yolov5權重比較==<div class="flex-container"><img src="https://i.imgur.com/RlOHVgT.jpg" width="100%"></div> ## 參、連接Firebase ### (一)安裝Firebase 1.打開PyCharm後在yolov5-master專案的Termial輸入： ``` pip install firebase ``` 2.接著輸入： ``` pip install --upgrade firebase-admin ``` ### (二)產生金鑰 1.進到Firebase後找到以下畫面 <div class="flex-container"><img src="https://i.imgur.com/gVy6viQ.png" width="100%"></div> 2.產生新的私密金鑰 <div class="flex-container"><img src="https://i.imgur.com/cPZk2vj.png" width="100%"></div> 3.將剛才下載好的金鑰檔案改名為serviceAccount後，放在yolov5專案資料夾下 <div class="flex-container"><img src="https://i.imgur.com/xiq6Uxc.png" width="100%"></div> ### (三)程式碼 1.複製資料庫網址 <div class="flex-container"><img src="https://i.imgur.com/Ew6lVHo.png" width="100%"></div> 2.前置作業: ```python= import firebase_admin from firebase_admin import db cred_obj = firebase_admin.credentials.Certificate('serviceAccount.json') default_app = firebase_admin.initialize_app(cred_obj, {'databaseURL': '資料庫網址'}) ``` 3.上傳資料: ```python= ref1 = db.reference("/movement/JennyPose/") #("/目錄名稱/節點1名稱/") ref1.set({ "times": '2021/10/28 21:54:10', "movement": 'Standing', "name": 'Jenny' }) #"資料標題":'資料' ref2 = db.reference("/movement/SophiaPose/") #("/目錄名稱/節點2名稱/") ref2.set({ "times": '2021/10/28 21:54:10', "movement": 'Sitting', "name": 'Sophia' }) #"資料標題":'資料' ``` 4.上傳結果(資料成功上傳至firebase裡) <div class="flex-container"><img src="https://i.imgur.com/D4IdHhY.png" width="100%"></div> ## 肆、完整程式碼 ```python= """Run inference with a YOLOv5 model on images, videos, directories, streams Usage: $ python path/to/detect.py --source path/to/img.jpg --weights yolov5s.pt --img 640 """ import argparse import sys import time from pathlib import Path import cv2 import torch import torch.backends.cudnn as cudnn FILE = Path(__file__).absolute() sys.path.append(FILE.parents[0].as_posix()) # add yolov5/ to path from models.experimental import attempt_load from utils.datasets import LoadStreams, LoadImages from utils.general import check_img_size, check_requirements, check_imshow, colorstr, non_max_suppression, \ apply_classifier, scale_coords, xyxy2xywh, strip_optimizer, set_logging, increment_path, save_one_box from utils.plots import colors, plot_one_box from utils.torch_utils import select_device, load_classifier, time_synchronized import json import requests import firebase_admin from firebase_admin import db #匯入firebase金鑰 cred_obj = firebase_admin.credentials.Certificate('serviceAccount.json') default_app = firebase_admin.initialize_app(cred_obj, {'databaseURL': 'https://elderlycare-f9022-default-rtdb.firebaseio.com/'}) #動作辨識一個動作的持續時間 AngelTime = [] SophiaTime = [] JennyTime = [] JimTime = [] TinaTime = [] @torch.no_grad() def run(weights='yolov5m.pt', # model.pt path(s) source='data/images', # file/dir/URL/glob, 0 for webcam imgsz=640, # inference size (pixels) conf_thres=0.25, # confidence threshold iou_thres=0.45, # NMS IOU threshold max_det=1000, # maximum detections per image device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu view_img=False, # show results save_txt=False, # save results to *.txt save_conf=False, # save confidences in --save-txt labels save_crop=False, # save cropped prediction boxes nosave=False, # do not save images/videos classes=None, # filter by class: --class 0, or --class 0 2 3 agnostic_nms=False, # class-agnostic NMS augment=False, # augmented inference update=False, # update all models project='runs/detect', # save results to project/name name='exp', # save results to project/name exist_ok=False, # existing project/name ok, do not increment line_thickness=3, # bounding box thickness (pixels) hide_labels=False, # hide labels hide_conf=False, # hide confidences half=False, # use FP16 half-precision inference ): save_img = not nosave and not source.endswith('.txt') # save inference images 是否保存图片 # 判断预测源是否为视频流 webcam = source.isnumeric() or source.endswith('.txt') or source.lower().startswith( ('rtsp://', 'rtmp://', 'http://', 'https://')) # Directories 获取保存预测路径 save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir # Initialize set_logging() # 初始化logging device = select_device(device) # 获取设备 half &= device.type != 'cpu' # half precision only supported on CUDA 如果设备为gpu且opt.half=True，使用Float16 # Load model model = attempt_load(weights, map_location=device) # load FP32 model 加载Float32模型，确保用户设定的输入图片分辨率能整除32(如不能则调整为能整除并返回) stride = int(model.stride.max()) # model stride imgsz = check_img_size(imgsz, s=stride) # check image size names = model.module.names if hasattr(model, 'module') else model.names # get class names 获取类别名字 if half: model.half() # to FP16 设置Float16 # Second-stage classifier 设置第二次分类，默认不使用 classify = False if classify: modelc = load_classifier(name='resnet50', n=2) # initialize modelc.load_state_dict(torch.load('resnet50.pt', map_location=device)['model']).to(device).eval() # Set Dataloader 通过不同的输入源来设置不同的数据加载方式 vid_path, vid_writer = None, None if webcam: # 检查当前环境是否能够正常imshow view_img = check_imshow() cudnn.benchmark = True # set True to speed up constant image size inference dataset = LoadStreams(source, img_size=imgsz, stride=stride) else: dataset = LoadImages(source, img_size=imgsz, stride=stride) # Run inference 进行一次前向推理,测试程序是否正常 if device.type != 'cpu': model(torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(next(model.parameters())))# run once t0 = time.time() ''' path 图片/视频路径 img 进行resize+pad之后的图片 img0 原size图片 cap 当读取图片时为None，读取视频时为视频源 ''' for path, img, im0s, vid_cap in dataset: img = torch.from_numpy(img).to(device) img = img.half() if half else img.float() # uint8 to fp16/32 图片也设置为Float16 img /= 255.0 # 0 - 255 to 0.0 - 1.0 # 没有batch_size的话则在最前面添加一个轴 if img.ndimension() == 3: img = img.unsqueeze(0) # Inference ''' 前向传播返回pred的shape是(1, num_boxes, 5+num_class) h,w为传入网络图片的长和宽，注意dataset在检测时使用了矩形推理，所以这里h不一定等于w num_boxes = h/32 * w/32 + h/16 * w/16 + h/8 * w/8 pred[..., 0:4]为预测框坐标预测框坐标为xywh(中心点+宽长)格式 pred[..., 4]为objectness置信度 pred[..., 5:-1]为分类结果 ''' t1 = time_synchronized() pred = model(img, augment=augment)[0] # Apply NMS ''' pred:前向传播的输出 conf_thres:置信度阈值 iou_thres:iou阈值 classes:是否只保留特定的类别 agnostic_nms:进行nms是否也去除不同类别之间的框 max-det:保留的最大检测框数量经过nms之后，预测框格式：xywh-->xyxy(左上角右下角) pred是一个列表list[torch.tensor]，长度为batch_size 每一个torch.tensor的shape为(num_boxes, 6),内容为box+conf+cls ''' pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det) t2 = time_synchronized() # Apply Classifier 添加二次分类，默认不使用 if classify: pred = apply_classifier(pred, modelc, img, im0s) # Process detections 对每一张图片作处理 # i代表enumerate作出來的前綴編號，det是pred裡面本身的內容 for i, det in enumerate(pred): # detections per image # 如果输入源是webcam，则batch_size不为1，取出dataset中的一张图片 if webcam: # batch_size >= 1 p, s, im0, frame = path[i], f'{i}: ', im0s[i].copy(), dataset.count else: p, s, im0, frame = path, '', im0s.copy(), getattr(dataset, 'frame', 0) p = Path(p) # to Path save_path = str(save_dir / p.name) # img.jpg 设置保存图片/视频的路径 txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # img.txt 设置保存框坐标txt文件的路径 s += '%gx%g ' % img.shape[2:] # print string 设置打印信息(图片长宽) gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh imc = im0.copy() if save_crop else im0 # for save_crop sent = ['non', 'non', 'non', 'non', 'non'] pointName = [] #所有辨識到的臉部的座標，按照x,y,x,y順序排列 pointMovement = [] #所有辨識到的動作的座標，按照x,y,x,y順序排列 postures = [] #所有辨識到的臉部的名稱 people = [] #所有辨識到的動作的名稱 if len(det): # Rescale boxes from img_size to im0 size #调整预测框的坐标：基于resize+pad的图片的坐标-->基于原size图片的坐标,此时坐标格式为xyxy det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round() # Print results 打印检测到的类别数量 for c in det[:, -1].unique(): #-1代表最後一個/行 n = (det[:, -1] == c).sum() # detections per class s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string ''' n為同一個標籤被辨識出的數量動作存入postures陣列臉部存入people陣列 ''' if (n==1): if (names[int(c)]) == "Sitting": postures.append("Sitting") elif (names[int(c)]) == "Standing": postures.append("Standing") elif (names[int(c)]) == "Lying Down": postures.append("Lying Down") elif (names[int(c)]) == "Angel": people.append("Angel") elif (names[int(c)]) == "Sophia": people.append("Sophia") elif (names[int(c)]) == "Jenny": people.append("Jenny") elif (names[int(c)]) == "Jim": people.append("Jim") elif (names[int(c)]) == "Tina": people.append("Tina") else: posture = [] person = [] if (names[int(c)]) == "Sitting": posture.append("Sitting") postures = posture * n elif (names[int(c)]) == "Standing": posture.append("Standing") postures=posture*n elif (names[int(c)]) == "Lying Down": posture.append("Lying Down") postures = posture * n elif (names[int(c)]) == "Angel": person.append("Angel") people = person * n elif (names[int(c)]) == "Sophia": person.append("Sophia") people = person * n elif (names[int(c)]) == "Jenny": person.append("Jenny") people = person * n elif (names[int(c)]) == "Jim": person.append("Jim") people = person * n elif (names[int(c)]) == "Tina": person.append("Tina") people = person * n # Write results 保存预测结果 for *xyxy, conf, cls in reversed(det): if save_txt: # Write to file #将xyxy(左上角+右下角)格式转为xywh(中心点+宽长)格式，并除上w，h做归一化，转化为列表再保存 xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh 1,4 line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format with open(txt_path + '.txt', 'a') as f: f.write(('%g ' * len(line)).rstrip() % line + '\n') if save_img or save_crop or view_img: # Add bbox to image 在原图上画框 c = int(cls) # integer class label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}') plot_one_box(xyxy, im0, label=label, color=colors(c, True), line_thickness=line_thickness) if save_crop: save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True) #保存crop centerWH = (xyxy2xywh(torch.tensor(xyxy).view(1, 4))).view(-1).tolist() #[中心點x座標，中心點y座標，寬，長] point = [centerWH[0], centerWH[1]] #[中心點x座標，中心點y座標] ''' 動作座標存入pointMovement陣列臉部座標存入pointName陣列 ''' if (names[int(c)]) == "Angel": pointName.append(point[0]) pointName.append(point[1]) elif (names[int(c)]) == "Sophia": pointName.append(point[0]) pointName.append(point[1]) elif (names[int(c)]) == "Jenny": pointName.append(point[0]) pointName.append(point[1]) elif (names[int(c)]) == "Jim": pointName.append(point[0]) pointName.append(point[1]) elif (names[int(c)]) == "Tina": pointName.append(point[0]) pointName.append(point[1]) elif (names[int(c)]) == "Sitting": pointMovement.append(point[0]) pointMovement.append(point[1]) elif (names[int(c)]) == "Standing": pointMovement.append(point[0]) pointMovement.append(point[1]) elif (names[int(c)]) == "Lying Down": pointMovement.append(point[0]) pointMovement.append(point[1]) pNX=pointName[0::2] #提出陣列pointName[0][2][4][6][8]位的值(辨識到的所有人臉的x座標) pNY=pointName[1::2] #提出陣列pointName[1][3][5][7][9]位的值(辨識到的所有人臉的y座標) pMX=pointMovement[0::2] #提出陣列pointMovement[0][2][4][6][8]位的值(辨識到的所有動作的x座標) pMY=pointMovement[1::2] #提出陣列pointMovement[1][3][5][7][9]位的值(辨識到的所有動作的y座標) nowTime = int(time.time()) # 取得現在時間 struct_time = time.localtime(nowTime) # 轉換成時間元組 timeString = time.strftime('%Y/%m/%d %H:%M:%S', struct_time) # 將時間元組轉換成想要的字串 # 計算時間 def timeCount(): if people[0] == "Angel": if postures[0] == "Sitting": if (not AngelTime) or (AngelTime[0]== "Sitting"): AngelTime.append("Sitting") AngelTime.append(nowTime) AngelTime_c = AngelTime[-1] - AngelTime[1] print(f'Angel Sitting Time {AngelTime_c} s') if (AngelTime[0] == "Standing") or (AngelTime[0] == "Lying Down"): AngelTime.clear() AngelTime.append("Sitting") AngelTime.append(nowTime) AngelTime_c = AngelTime[-1] - AngelTime[1] print(f'Angel Sitting Time {AngelTime_c} s') if postures[0] == "Standing": if (not AngelTime) or (AngelTime[0] == "Standing"): AngelTime.append("Standing") AngelTime.append(nowTime) AngelTime_c = AngelTime[-1] - AngelTime[1] print(f'Angel Standing Time {AngelTime_c} s') if (AngelTime[0] == "Sitting") or (AngelTime[0] == "Lying Down"): AngelTime.clear() AngelTime.append("Standing") AngelTime.append(nowTime) AngelTime_c = AngelTime[-1] - AngelTime[1] print(f'Angel Standing Time {AngelTime_c} s') if postures[0] == "Lying Down": if (not AngelTime) or (AngelTime[0] == "Lying Down"): AngelTime.append("Lying Down") AngelTime.append(nowTime) AngelTime_c = AngelTime[-1] - AngelTime[1] print(f'Angel Lying Down Time {AngelTime_c} s') if (AngelTime[0] == "Standing") or (AngelTime[0] == "Sitting"): AngelTime.clear() AngelTime.append("Lying Down") AngelTime.append(nowTime) AngelTime_c = AngelTime[-1] - AngelTime[1] print(f'Angel Lying Down Time {AngelTime_c} s') if people[0] == "Sophia": if postures[0] == "Sitting": if (not SophiaTime) or (SophiaTime[0]== "Sitting"): #判斷陣列為空，或是陣列第一筆為Sitting SophiaTime.append("Sitting") #存入時間前先存入一個text，用來做判斷是否有換動作 SophiaTime.append(nowTime) #將時間存入陣列 SophiaTime_c = SophiaTime[-1] - SophiaTime[1] #最後一個時間 - 第一個時間 print(f'Sophia Sitting Time {SophiaTime_c} s') if (SophiaTime_c > 2): print("fcm start") fcm() #測試傳fcm if (SophiaTime[0] == "Standing") or (SophiaTime[0] == "Lying Down"): #假如陣列第一筆不是Sitting SophiaTime.clear() #清空陣列 SophiaTime.append("Sitting") SophiaTime.append(nowTime) SophiaTime_c = SophiaTime[-1] - SophiaTime[1] print(f'Sophia Sitting Time {SophiaTime_c} s') if (SophiaTime_c > 2): print("fcm start") fcm() if postures[0] == "Standing": if (not SophiaTime) or (SophiaTime[0] == "Standing"): SophiaTime.append("Standing") SophiaTime.append(nowTime) SophiaTime_c = SophiaTime[-1] - SophiaTime[1] print(f'Sophia Standing Time {SophiaTime_c} s') if (SophiaTime[0] == "Sitting") or (SophiaTime[0] == "Lying Down"): SophiaTime.clear() SophiaTime.append("Standing") SophiaTime.append(nowTime) SophiaTime_c = SophiaTime[-1] - SophiaTime[1] print(f'Sophia Standing Time {SophiaTime_c} s') if postures[0] == "Lying Down": if (not SophiaTime) or (SophiaTime[0] == "Lying Down"): SophiaTime.append("Lying Down") SophiaTime.append(nowTime) SophiaTime_c = SophiaTime[-1] - SophiaTime[1] print(f'Sophia Lying Down Time {SophiaTime_c} s') if (SophiaTime[0] == "Standing") or (SophiaTime[0] == "Sitting"): SophiaTime.clear() SophiaTime.append("Lying Down") SophiaTime.append(nowTime) SophiaTime_c = SophiaTime[-1] - SophiaTime[1] print(f'Sophia Lying Down Time {SophiaTime_c} s') if people[0] == "Jenny": if postures[0] == "Sitting": if (not JennyTime) or (JennyTime[0]== "Sitting"): JennyTime.append("Sitting") JennyTime.append(nowTime) JennyTime_c = JennyTime[-1] - JennyTime[1] print(f'Jenny Sitting Time {JennyTime_c} s') if (JennyTime[0] == "Standing") or (JennyTime[0] == "Lying Down"): JennyTime.clear() JennyTime.append("Sitting") JennyTime.append(nowTime) JennyTime_c = JennyTime[-1] - JennyTime[1] print(f'Jenny Sitting Time {JennyTime_c} s') if postures[0] == "Standing": if (not JennyTime) or (JennyTime[0] == "Standing"): JennyTime.append("Standing") JennyTime.append(nowTime) JennyTime_c = JennyTime[-1] - JennyTime[1] print(f'Jenny Standing Time {JennyTime_c} s') if (JennyTime[0] == "Sitting") or (JennyTime[0] == "Lying Down"): JennyTime.clear() JennyTime.append("Standing") JennyTime.append(nowTime) JennyTime_c = JennyTime[-1] - JennyTime[1] print(f'Jenny Standing Time {JennyTime_c} s') if postures[0] == "Lying Down": if (not JennyTime) or (JennyTime[0] == "Lying Down"): JennyTime.append("Lying Down") JennyTime.append(nowTime) JennyTime_c = JennyTime[-1] - JennyTime[1] print(f'Jenny Lying Down Time {JennyTime_c} s') if (JennyTime[0] == "Standing") or (JennyTime[0] == "Sitting"): JennyTime.clear() JennyTime.append("Lying Down") JennyTime.append(nowTime) JennyTime_c = JennyTime[-1] - JennyTime[1] print(f'Jenny Lying Down Time {JennyTime_c} s') if people[0] == "Jim": if postures[0] == "Sitting": if (not JimTime) or (JimTime[0]== "Sitting"): JimTime.append("Sitting") JimTime.append(nowTime) JimTime_c = JimTime[-1] - JimTime[1] print(f'Jim Sitting Time {JimTime_c} s') if (JimTime[0] == "Standing") or (JimTime[0] == "Lying Down"): JimTime.clear() JimTime.append("Sitting") JimTime.append(nowTime) JimTime_c = JimTime[-1] - JimTime[1] print(f'Jim Sitting Time {JimTime_c} s') if postures[0] == "Standing": if (not JimTime) or (JimTime[0] == "Standing"): JimTime.append("Standing") JimTime.append(nowTime) JimTime_c = JimTime[-1] - JimTime[1] print(f'Jim Standing Time {JimTime_c} s') if (JimTime[0] == "Sitting") or (JimTime[0] == "Lying Down"): JimTime.clear() JimTime.append("Standing") JimTime.append(nowTime) JimTime_c = JimTime[-1] - JimTime[1] print(f'Jim Standing Time {JimTime_c} s') if postures[0] == "Lying Down": if (not JimTime) or (JimTime[0] == "Lying Down"): JimTime.append("Lying Down") JimTime.append(nowTime) JimTime_c = JimTime[-1] - JimTime[1] print(f'Jim Lying Down Time {JimTime_c} s') if (JimTime[0] == "Standing") or (JimTime[0] == "Sitting"): JimTime.clear() JimTime.append("Lying Down") JimTime.append(nowTime) JimTime_c = JimTime[-1] - JimTime[1] print(f'Jim Lying Down Time {JimTime_c} s') if people[0] == "Tina": if postures[0] == "Sitting": if (not TinaTime) or (TinaTime[0]== "Sitting"): TinaTime.append("Sitting") TinaTime.append(nowTime) TinaTime_c = TinaTime[-1] - TinaTime[1] print(f'Tina Sitting Time {TinaTime_c} s') if (TinaTime[0] == "Standing") or (TinaTime[0] == "Lying Down"): TinaTime.clear() TinaTime.append("Sitting") TinaTime.append(nowTime) TinaTime_c = TinaTime[-1] - TinaTime[1] print(f'Tina Sitting Time {TinaTime_c} s') if postures[0] == "Standing": if (not TinaTime) or (TinaTime[0] == "Standing"): TinaTime.append("Standing") TinaTime.append(nowTime) TinaTime_c = TinaTime[-1] - TinaTime[1] print(f'Tina Standing Time {TinaTime_c} s') if (TinaTime[0] == "Sitting") or (TinaTime[0] == "Lying Down"): TinaTime.clear() TinaTime.append("Standing") TinaTime.append(nowTime) TinaTime_c = TinaTime[-1] - TinaTime[1] print(f'Tina Standing Time {TinaTime_c} s') if postures[0] == "Lying Down": if (not TinaTime) or (TinaTime[0] == "Lying Down"): TinaTime.append("Lying Down") TinaTime.append(nowTime) TinaTime_c = TinaTime[-1] - TinaTime[1] print(f'Tina Lying Down Time {TinaTime_c} s') if (TinaTime[0] == "Standing") or (TinaTime[0] == "Sitting"): TinaTime.clear() TinaTime.append("Lying Down") TinaTime.append(nowTime) TinaTime_c = TinaTime[-1] - TinaTime[1] print(f'Tina Lying Down Time {TinaTime_c} s') def dataUpdateFirebase(): timeCount() #呼叫計算時間的方法，同時上傳資料同時計算時間 data1 = {'Object': ([people[0]], [postures[0]])} data = {'name': people[0], 'movement': postures[0], 'times': timeString} if sent[0] == 'non': if data1['Object'] == (['Angel'], ['Sitting']): sent[0] = 'Sitting' print(data) ref0 = db.reference("/movement/E001AngelPose/") #資料上傳firebase要存入的節點 ref0.set({ #要上傳的資料 "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) #將人臉x座標、y座標，動作x座標、y座標，人臉陣列、動作陣列的第一筆資料刪除 elif data1['Object'] == (['Angel'], ['Standing']): sent[0] = 'Standing' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Angel'], ['Lying Down']): sent[0] = 'Lying Down' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[0] != 'non': if sent[0] == 'Sitting': if data1['Object'] == (['Angel'], ['Standing']): sent[0] = 'Standing' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Angel'], ['Lying Down']): sent[0] = 'Lying Down' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[0] == 'Standing': if data1['Object'] == (['Angel'], ['Sitting']): sent[0] = 'Sitting' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Angel'], ['Lying Down']): sent[0] = 'Lying Down' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[0] == 'Lying Down': if data1['Object'] == (['Angel'], ['Sitting']): sent[0] = 'Sitting' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Angel'], ['Standing']): sent[0] = 'Standing' print(data) ref0 = db.reference("/movement/E001AngelPose/") ref0.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E001" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) if sent[1] == 'non': if data1['Object'] == (['Sophia'], ['Sitting']): sent[1] = 'Sitting' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Sophia'], ['Standing']): sent[1] = 'Standing' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Sophia'], ['Lying Down']): sent[1] = 'Lying Down' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[1] != 'non': if sent[1] == 'Sitting': if data1['Object'] == (['Sophia'], ['Standing']): sent[1] = 'Standing' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Sophia'], ['Lying Down']): sent[1] = 'Lying Down' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[1] == 'Standing': if data1['Object'] == (['Sophia'], ['Sitting']): sent[1] = 'Sitting' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Sophia'], ['Lying Down']): sent[1] = 'Lying Down' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[1] == 'Lying Down': if data1['Object'] == (['Sophia'], ['Sitting']): sent[1] = 'Sitting' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Sophia'], ['Standing']): sent[1] = 'Standing' print(data) ref1 = db.reference("/movement/E002SophiaPose/") ref1.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E002" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) if sent[2] == 'non': if data1['Object'] == (['Jenny'], ['Sitting']): sent[2] = 'Sitting' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jenny'], ['Standing']): sent[2] = 'Standing' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jenny'], ['Lying Down']): sent[2] = 'Lying Down' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[2] != 'non': if sent[2] == 'Sitting': if data1['Object'] == (['Jenny'], ['Standing']): sent[2] = 'Standing' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jenny'], ['Lying Down']): sent[2] = 'Lying Down' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[2] == 'Standing': if data1['Object'] == (['Jenny'], ['Sitting']): sent[2] = 'Sitting' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jenny'], ['Lying Down']): sent[2] = 'Lying Down' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[2] == 'Lying Down': if data1['Object'] == (['Jenny'], ['Sitting']): sent[2] = 'Sitting' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jenny'], ['Standing']): sent[2] = 'Standing' print(data) ref2 = db.reference("/movement/E003JennyPose/") ref2.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E003" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) if sent[3] == 'non': if data1['Object'] == (['Jim'], ['Sitting']): sent[3] = 'Sitting' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jim'], ['Standing']): sent[3] = 'Standing' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jim'], ['Lying Down']): sent[3] = 'Lying Down' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[3] != 'non': if sent[3] == 'Sitting': if data1['Object'] == (['Jim'], ['Standing']): sent[3] = 'Standing' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jim'], ['Lying Down']): sent[3] = 'Lying Down' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[3] == 'Standing': if data1['Object'] == (['Jim'], ['Sitting']): sent[3] = 'Sitting' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jim'], ['Lying Down']): sent[3] = 'Lying Down' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[3] == 'Lying Down': if data1['Object'] == (['Jim'], ['Sitting']): sent[3] = 'Sitting' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Jim'], ['Standing']): sent[3] = 'Standing' print(data) ref3 = db.reference("/movement/E004JimPose/") ref3.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E004" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) if sent[4] == 'non': if data1['Object'] == (['Tina'], ['Sitting']): sent[4] = 'Sitting' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Tina'], ['Standing']): sent[4] = 'Standing' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Tina'], ['Lying Down']): sent[4] = 'Lying Down' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[4] != 'non': if sent[4] == 'Sitting': if data1['Object'] == (['Tina'], ['Standing']): sent[4] = 'Standing' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Tina'], ['Lying Down']): sent[4] = 'Lying Down' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[4] == 'Standing': if data1['Object'] == (['Tina'], ['Sitting']): sent[4] = 'Sitting' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Tina'], ['Lying Down']): sent[4] = 'Lying Down' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif sent[4] == 'Lying Down': if data1['Object'] == (['Tina'], ['Sitting']): sent[4] = 'Sitting' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) elif data1['Object'] == (['Tina'], ['Standing']): sent[4] = 'Standing' print(data) ref4 = db.reference("/movement/E005TinaPose/") ref4.set({ "times": timeString, "movement": postures[0], "name": people[0], "ID": "E005" }) print("已上傳") del (pNX[0], pNY[0], pMX[0], pMY[0], people[0], postures[0]) j = len(people) #people陣列的長度 for i in range(j): if (len(postures) >= 1)&(len(people) >= 1): lenX = abs(pNX[0] - pMX[0]) #取絕對值(人臉x座標 - 動作x座標) lenY = abs(pNY[0] - pMY[0]) #取絕對值(人臉y座標 - 動作y座標) if (lenX >= 0.0) & (lenX <= 200.0) & (lenY >= 0.0) & (lenY <= 2000.0): dataUpdateFirebase() #呼叫將資料上傳firebase的方法 #傳fcm def SendMessageToLineNotify(Message, MessagingSenderId, Token): headers = {"content-type": "application/json","Authorization": "key=" + MessagingSenderId} url = "https://fcm.googleapis.com/fcm/send" data = {"notification": {"title" : "警訊!", "body" : Message}, "to" : Token} req = requests.post(url, data=json.dumps(data), headers=headers) print (req.text) def fcm(): Message = "躺著超過九小時!" MessagingSenderId = "1:326497554126:android:d2f4d30ad71005a8630a9f" #應用程式ID #MessagingSenderId = "326497554126" #寄件者ID #fcm 伺服器金鑰 #MessagingSenderId = "AAAATATFTs4:APA91bFztQHHI9sEOg4QY_jmQo9d4DjkZtQcg4u-_USb3Yc62eWojQmHEejyPd3X0jseWgaINEsZnOH6B24Q3DmhC01qtsWcsgIIawqKSy25gNwWhY82iprYGfu0mzzFGsmElw3lOyk4" # fcm 伺服器金鑰 Token = "AAAATATFTs4:APA91bFztQHHI9sEOg4QY_jmQo9d4DjkZtQcg4u-_USb3Yc62eWojQmHEejyPd3X0jseWgaINEsZnOH6B24Q3DmhC01qtsWcsgIIawqKSy25gNwWhY82iprYGfu0mzzFGsmElw3lOyk4" #Token = "1:326497554126:android:d2f4d30ad71005a8630a9f" #應用程式ID #Token = "FHf7Bi5mHgPqEMeZULqF6KReW7p2" #使用者UID SendMessageToLineNotify(Message, MessagingSenderId, Token) # Print time (inference + NMS) 打印前向传播+nms时间 print(f'{s}Done. ({t2 - t1:.3f}s)') # Stream results 如果设置展示，则show图片/视频 if view_img: cv2.imshow(str(p), im0) cv2.waitKey(1) # 1 millisecond # Save results (image with detections) 设置保存图片/视频 if save_img: if dataset.mode == 'image': cv2.imwrite(save_path, im0) else: # 'video' or 'stream' if vid_path != save_path: # new video vid_path = save_path if isinstance(vid_writer, cv2.VideoWriter): vid_writer.release() # release previous video writer if vid_cap: # video fps = vid_cap.get(cv2.CAP_PROP_FPS) w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH)) h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) else: # stream fps, w, h = 30, im0.shape[1], im0.shape[0] save_path += '.mp4' vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h)) vid_writer.write(im0) #显示保存信息 if save_txt or save_img: s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else '' print(f"Results saved to {save_dir}{s}") #strip_optimizer函数将pt文件中除了模型model或者ema之外的所有东西去除 if update: strip_optimizer(weights) # update model (to fix SourceChangeWarning) print(f'Done. ({time.time() - t0:.3f}s)') def parse_opt(): parser = argparse.ArgumentParser() parser.add_argument('--weights', nargs='+', type=str, default='bestWeights/best3.pt', help='model.pt path(s)') # 選用訓練的權重，可用根目錄下的yolov5s.pt，也可用runs/train/exp/weights/best.pt #parser.add_argument('--weights', nargs='+', type=str, default=['bestAct.pt','bestAct2.pt'], help='model.pt path(s)') # 選用兩個訓練的權重 parser.add_argument('--source', type=str, default='data/ElderlyCareTest', help='file/dir/URL/glob, 0 for webcam') # 檢測資料，可以是圖片/視訊路徑，也可以是'0'(電腦自帶攝像頭),也可以是rtsp等視訊流 #parser.add_argument('--source', type=str, default='0', help='file/dir/URL/glob, 0 for webcam')#連接webcam parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='inference size (pixels)') # 網路輸入圖片大小 parser.add_argument('--conf-thres', type=float, default=0.25, help='confidence threshold') # 置信度閾值，檢測到的物件屬於特定類（狗，貓，香蕉，汽車等）的概率 parser.add_argument('--iou-thres', type=float, default=0.45, help='NMS IoU threshold') # 做nms的iou閾值 parser.add_argument('--max-det', type=int, default=1280, help='maximum detections per image') #保留的最大检测框数量 parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu') # 檢測的裝置，cpu；0(表示一個gpu裝置cuda:0)；0,1,2,3(多個gpu裝置)。值為空時，訓練時預設使用計算機自帶的顯示卡或CPU parser.add_argument('--view-img', action='store_true', help='show results') # 是否展示檢測之後的圖片/視訊，預設False parser.add_argument('--save-txt', action='store_true', help='save results to *.txt') # 是否將檢測的框座標以txt檔案形式儲存，預設False parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels') # 是否將檢測的labels以txt檔案形式儲存，預設False parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes') #是否保存裁剪预测框图片 parser.add_argument('--nosave', action='store_true', help='do not save images/videos') #不保存图片、视频 parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3') # 設定只保留某一部分類別，如0或者0 2 3 parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS') # 進行nms是否也去除不同類別之間的框，預設False parser.add_argument('--augment', action='store_true', help='augmented inference') # 推理的時候進行多尺度，翻轉等操作(TTA)推理 parser.add_argument('--update', action='store_true', help='update all models') # 如果為True，則對所有模型進行strip_optimizer操作，去除pt檔案中的優化器等資訊，預設為False parser.add_argument('--project', default='runs/ElderlyCare', help='save results to project/name') # 檢測結果所存放的路徑，預設為runs/detect parser.add_argument('--name', default='expElderlyCare', help='save results to project/name') # 檢測結果所在資料夾的名稱，預設為exp parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment') # 若現有的project/name存在，則不進行遞增 parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)') #画框的线条粗细 parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels') #可视化时隐藏预测类别 parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences') #可视化时隐藏置信度 parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference') #是否使用F16精度推理 opt = parser.parse_args() return opt def main(opt): print(colorstr('ElderlyCare: ') + ', '.join(f'{k}={v}' for k, v in vars(opt).items())) # 检查环境 check_requirements(exclude=('tensorboard', 'thop')) run(**vars(opt)) if __name__ == "__main__": opt = parse_opt() main(opt) ``` ## 參考資料 (環境安裝建置) https://blog.csdn.net/qq_44697805/article/details/107702939 https://tw511.com/a/01/13004.html https://www.uj5u.com/houduan/277337.html (YoloV4環境建置，用來參考Cuda、Cudnn安裝方式) https://www.youtube.com/watch?v=PVf16gIhnek (Cuda GPU支援狀況，數字越大，運算越快，原則上3以上就沒問題了) https://en.wikipedia.org/wiki/CUDA (訓練自製數據參考資料) https://tw511.com/a/01/29504.html https://blog.csdn.net/g11d111/article/details/108872076 https://www.uj5u.com/houduan/277337.html (利用pycocotools的whl檔安裝方式) https://iter01.com/561896.html (apex安裝方式) https://blog.csdn.net/mrjkzhangma/article/details/100704397 (精靈標記助手(Colabeler)使用) https://blog.csdn.net/youmumzcs/article/details/79657132 (YoloV5簡介) https://zhuanlan.zhihu.com/p/172121380 (Pycharm連接Firebase參考資料) https://medium.com/%E7%A8%8B%E5%BC%8F%E8%A3%A1%E6%9C%89%E8%9F%B2/python-%E7%88%AC%E8%9F%B2%E4%B8%8D%E6%B1%82%E4%BA%BA%E4%B9%8B%E8%B3%87%E6%96%99%E5%84%B2%E5%AD%98%E7%AF%87-9bc0146e56f1 https://kk665403.pixnet.net/blog/post/403982819-%5Bpython%5D-python-firebase%E8%B3%87%E6%96%99%E5%BA%AB%E4%B8%B2%E6%8E%A5%E6%93%8D%E4%BD%9C%E7%B0%A1%E6%98%93%E6%96%B9%E6%B3%95 https://kk665403.pixnet.net/blog/post/403982819-%5Bpython%5D-python-firebase%E8%B3%87%E6%96%99%E5%BA%AB%E4%B8%B2%E6%8E%A5%E6%93%8D%E4%BD%9C%E7%B0%A1%E6%98%93%E6%96%B9%E6%B3%95