# 2022 IOT # 路邊停車格即時資訊 ##### Real-time On-street Parking System for Scooters and Motorcycle 組員： 施育衡，周宗翰，劉柏顯，陳捷文，陳威宇 --- # Motivation ---- <!-- 台灣的交通主要由機車所構成，並且多數機車族最大的煩惱莫過於尋找停車位，路邊的機車位成了他們的首選。 --> In Taiwan, the traffic is mainly consisted of scooters and motorcycles. The biggest problem for them is finding a space to park their vehicle. For convenience, on-street parking space becomes their first choice. ---- <!-- 但是路邊的停車格與有管理的停車場不同，沒辦法很好的追蹤停車數量。 --> The biggest difference between on-street parking space and managed parking lot is admission control. It is difficult to track the amount of vehicles parked for on-street parking space. ---- <!-- 因此使用者無法預測此路段何時會有路邊車位，或此路段的停車壅擠程度。 --> Therefore, there isn't a well-designed system for tracking that whether a road section will have empty parking space or the congestion level of that on-street parking space. ---- <!-- 所以我們擬研發一套即時機車路邊停車系統， --> We consider developing a real-time on-street parking system to resolve this problem. ---- <!-- 方便使用者追蹤目的地附近車位，減少等待時間，進一步改善交通問題。 --> We aims to provide a system for user to track on-street parking space near their destination to minimize their waiting time, and improve overall traffic congestion. --- # Feature ---- <!-- 紀錄、分析台灣路邊機車停車格的車位資訊。 --> 1. Record and analyze parking space information of on-street parking space for Scooters/Motorcycle in Taiwan. ---- <!-- 提供使用者即時的車位資訊，並透過使用者的位置提供車位推薦。 --> 2. This system will provide real-time information to the user and recommend parking space with respect to users location. --- <!-- ## 商業價值 --> # Commercial Value ---- <!-- 分析使用者喜好，並結合附近展覽，活動，或美食店家等, 推薦哪一些停車空位附近是否有使用者感興趣的事物。 --> Analyzing user preferences and combining nearby exhibitions, events, or restaurant, and etc.. To recommend where parking spaces are nearby if there are things that users are interested in. ---- <!-- 另外，分析目標停車場附近的交通概況，加速人流流動，增加曝光率與遊客人數。 --> Analyze the traffic situation about the target parking place near destination, to accelerate the flow of people, and increase the exposure rate of exhibition. <!-- # 參考文獻 先放上來: --- 無人機: 1. [UAV assisted smart parking solution](https://www.researchgate.net/publication/319034631_UAV_assisted_smart_parking_solution) * [Summary（未完整版）](https://docs.google.com/document/d/1zVMGndJ0emlPzm6ueRbDHLNoH1NrcrffyHxOxu1RoOA/edit?usp=sharing) 2. [A Large Contextual Dataset for Classification, Detection and Counting of Cars with Deep Learning](https://paperswithcode.com/paper/a-large-contextual-dataset-for-classification) --- 相機: 3. [Car parking occupancy detection using smart camera networks and Deep Learning](https://www.researchgate.net/publication/306304601_Car_parking_occupancy_detection_using_smart_camera_networks_and_Deep_Learning) 4. [Real Time IP Camera Parking Occupancy Detection using Deep Learning](https://www.researchgate.net/publication/349486692_Real_Time_IP_Camera_Parking_Occupancy_Detection_using_Deep_Learning) --- 資料分析: 5. [IoT Based Smart Parking System Using Deep Long Short Memory Network](https://www.mdpi.com/2079-9292/9/10/1696/htm) --- 影像物體辨識: 6. [YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors](https://arxiv.org/abs/2207.02696) --- 伺服器系統 7. https://github.com/offenesdresden/ParkAPI --> --- <!-- # 參考文獻與討論 --> # Related Work and Discussion ---- ## Detection [YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors](https://arxiv.org/abs/2207.02696) <!-- Yolo是一個物體辨識的深度學習模型，給予一張影像，模型會辨識物體的類別與 bounding box 的位置。 --> Yolo is a deep learning model for object detection. Given a picture, Yolo will categorize the objects and frame out the position with bounding box for each object. <!-- 我們可以使用detection的方法經過辨識後，再對偵測後的label進行處理 --> We wish to design the system to use the detection methods to recognize vehicles inside each parking space, and then process the coresponding label. <!-- 優點: 可以同時汽車與摩托車的停車狀態 --> The benefit of using Yolo is that we can detect both car and scooter/motorcycle's parking status. <!-- 缺點: 若是影像過大或物體過於密集，此模型處理速度慢且性能不佳 --> The challenge of this method is that if the detection area is too large or objects are too dense, the model may spent more time on processing. ---- ## Counting [An accurate car counting in aerial images based on convolutional neural networks](https://link.springer.com/content/pdf/10.1007/s12652-021-03377-5.pdf) <!-- 透過輸入影像，輸出影像的density map，再透過積分的方式數出圖中關注的物體數量 --> This work aims to predict density map of a given image, and use integral methods to count the amount of relevent objects. <!-- 優點: 可以在物體密集及嚴重遮擋的情況下，很好地預測圖中的物體數量。 --> The benefit of this method is that it can detect objects even if its severly blocked and congested. <!-- 缺點: 無法精準定位與得知物體大小 --> The challenge of using this method is that it cannot precisely locate objects and predict its size. ---- ## Data Analysis [IoT Based Smart Parking System Using Deep Long Short Memory Network](https://www.mdpi.com/2079-9292/9/10/1696/htm) <!-- 平均來說每個司機要花 20 分鐘找停車位，相對的有 30% 車流堵塞是因為尋找車位造成的，為了減輕車流壅塞的問題，使用基於 LSTM 模型搭配時間序列去預測停車狀況，透過整合 IoT, cloud technology, and sensor networks 來，開發停車預測模擬系統，預測空車位，減少使用者等待時間，讓使用者可以更好的進行目的地規劃與行程 --> In average, it costs every car driver 20 minutes to find an empty parking space. Also, Statistics shows that 30% of traffic congestion are caused by finding empty parking spaces. To resolve this problem, this work develop a LSTM model and use time sequence to predict parking status. This work also integrate IoT, cloud technology and sensor network to construct a complete parking forecast and analysis system. --- <!-- # 問題與挑戰 --> # Problem and Challenges ---- <!--即時系統 需在短時間內更新資訊--> <!--路上障礙物干擾 偵測器辨識--> <!--尖峰時期相較大量使用者--> <!--錯誤車輛停在停車格上--> 1. The need for short delay real-time system 2. Interference to sensors by obstacles and extreme condition 3. Rapid data update due to relative mass flow of traffic 4. Wrong vehicle on parking slots 5. Project cost should be at a fair price ---- <!-- 1. 偵測範圍與設備價值比例 2. 分析停車時間、動向、使用者動機 3. 合理分配使用者至各至車位，不將過多使用者分配至同樣車位 4. 結合店商活動、讓使用者更有機會經過有興趣的店家 (使用者興趣、取向分析?) 5. 狹窄、老舊道路 6. 違停情況分析 --> 1. Trade off maximize area coverage and minimize cost. 2. Detect vehicle and user intent. 3. Dynamic parking slot management. 4. Commercial adverticement to user combined with parking. 5. Congested parking space analysis caused by illegal parking and insufficient space. 1,2,3 efficiency issue 4 cost issue 5,6,7 Errors processing due to some reason 8,9 commercial --- # Sensors ---- - Camera (sigular or multiple): Webcam or CCTV camera (differ from video quality and price) - Ultrasonic sensors - radar sensors ---