---
tags: 程式碼
---

# Seeing Red (程式碼結果)
> 追蹤 BPM_83.MOV (00015) 為例
> [論文筆記](https://hackmd.io/@ncnucsie/SymKoJ2a9)

### signal_extractor.py
==從 params.yaml 裡修改要套用進去的 func，程式預設是使用 luma_component_mean 和 red_chennel_mean<br><br>作者把每一幀的數值存進csv檔裡<br>==
<br>

### signal_preprocessor.py
==原程式預設兩條路線<br>==
> 之後可以再試其他組合

"hpf":       "butter_highpass_filter",
"lpf":       "butter_lowpass_filter",
"maf":       "moving_average_flat",
"diff_pad":  "minus_with_pad",
"fft":       "fft",
"roll_avg":  "rolling_average",
"sub":       "subtract",
"bandpass":  "butter_bandpass_filter",
"imf":       "increase_main_freq",
"cut_start": "cut_start",
"bpf_bpm":   "bandpass_bpm"

1. chain2: roll_avg -> sub -> low-pass filter -> cut_start<br>
> parameter:
> roll_avg -> window_size_seconds: 1.01
> low-pass filter -> filter_order: 2 / low: 4
> cut_start -> seconds: 3

2. dynamic_bpm: cut_start -> high-pass filter -> band-pass filter_bpm<br>
> parameter:
> cut_start -> seconds: 3
> high-pass filter -> cutoff: 0.5 / order: 1
> band-pass filter_bpm -> multiplier: 3 / mincut: 0.01 / order: 1


### signal_fiducial_points_detection.py
==套論文公式==
<br>

### signal_fiducial_points_detection.py
1. systolic peak -> 對曲線做一次微分，頂點的斜率為0 + 取最大值
2. dicrotic notch -> 一次微分值為負數 + 二次微分值接近0
3. diastolic peak -> 一次微分值接近0 + 二次微分值接近0
<br>

### 分析每個波適不適合當作分析資料
<br>