基於 OpenCV 人臉檢測和 POS 算法

• 🎯 自包含實現 - 不依賴有問題的外部庫，完全自主實現
• 👥 可靠的人臉檢測 - 使用 OpenCV Haar 級聯，支援多參數組合
• 💓 POS 算法 - Plane-Orthogonal-to-Skin 方法，提供穩健的心率估測
• 📊 即時進度更新 - Console 和 GUI 雙重進度顯示
• 🔍 人臉檢測測試 - 預先檢查影片是否適合分析
• 📈 資料匯出 - 下載心率圖表和 CSV 資料
• ⚡ 高效處理 - 分段處理，2分鐘影片約 30 秒完成

# 安裝依賴套件
pip install gradio opencv-python numpy scipy matplotlib tqdm

# 執行應用程式
python app.py

在瀏覽器中開啟 http://127.0.0.1:7860

1. 上傳包含清晰正面人臉的影片檔案
2. 調整參數（預設值通常已足夠）
3. 點擊「Process Video」開始分析
4. 查看結果並下載資料

1. 上傳影片進行人臉檢測測試
2. 點擊「Test Face Detection」
3. 確認影片是否適合進行心率分析

• 方法 (Method): POS (Plane-Orthogonal-to-Skin) - 目前唯一實現的方法
• 視窗 (Window): 心率計算的時間窗口 (5-30秒，預設: 10秒)
• 步長 (Step): 更新間隔 (0.5-5秒，預設: 2秒)
• 心率範圍 (Min/Max BPM): 心率過濾範圍 (預設: 45-180 BPM)
• 信心閾值 (Confidence Threshold): 有效讀數的最低信心值 (0.0-1.0，預設: 0.3)

• 清晰正面人臉: 人臉應清楚可見且面向攝影機
• 良好光線: 充足的光線，避免強烈陰影
• 最小移動: 避免過度的頭部移動或攝影機晃動
• 影片長度: 至少 30 秒以獲得可靠測量
• 解析度: 較高解析度通常效果更好

• 分段處理: 使用滑動窗口，避免逐幀處理
• 智能終止: 找到足夠測量值後自動停止
• 跳躍讀取: 直接跳到目標幀，提高效率
• 雙重進度: Console 和 GUI 進度條同步顯示

A simple and effective implementation for estimating heart rate from video using remote photoplethysmography (rPPG) with OpenCV face detection and POS algorithm.

• 🎯 Self-contained rPPG implementation - No dependency on problematic external libraries
• 👥 Reliable face detection - Uses OpenCV Haar cascades with multiple parameter sets
• 💓 POS algorithm - Plane-Orthogonal-to-Skin method for robust heart rate estimation
• 📊 Real-time progress updates - Console and GUI dual progress display
• 🔍 Face detection test - Pre-check videos for suitability before processing
• 📈 Export capabilities - Download heart rate plots and CSV data
• ⚡ Efficient processing - Chunk-based processing, ~30 seconds for 2-minute video

# Install dependencies
pip install gradio opencv-python numpy scipy matplotlib tqdm

# Run the application
python app.py

Access the web interface at http://127.0.0.1:7860

1. Upload a video file containing a clear frontal face
2. Adjust parameters if needed (defaults work well)
3. Click "Process Video" to start analysis
4. View results and download data

1. Upload a video to test face detection capability
2. Click "Test Face Detection"
3. Check if your video is suitable for heart rate analysis

• Method: POS (Plane-Orthogonal-to-Skin) - Currently the only implemented method
• Window: Time window for heart rate calculation (5-30 seconds, default: 10s)
• Step: Update interval (0.5-5 seconds, default: 2s)
• Min/Max BPM: Heart rate range filter (default: 45-180 BPM)
• Confidence Threshold: Minimum confidence for valid readings (0.0-1.0, default: 0.3)

• Clear frontal face: Face should be clearly visible and facing the camera
• Good lighting: Adequate lighting without harsh shadows
• Minimal movement: Avoid excessive head movement or camera shake
• Duration: At least 30 seconds for reliable measurements
• Resolution: Higher resolution generally works better

• Chunk-based processing: Uses sliding windows, avoids frame-by-frame processing
• Smart termination: Automatically stops after finding sufficient measurements
• Jump reading: Directly jumps to target frames for improved efficiency
• Dual progress: Console and GUI progress bars synchronized

• Face Detection: OpenCV Haar cascade classifiers with multiple parameter sets
• ROI Extraction: Forehead and cheek regions for optimal signal quality
• Signal Processing: POS algorithm with temporal bandpass filtering
• Heart Rate Estimation: FFT-based frequency domain analysis

1. Face Detection: Detect faces using OpenCV with fallback parameters
2. ROI Selection: Extract regions of interest from detected faces
3. Signal Extraction: Compute mean RGB values from ROI
4. POS Processing: Apply Plane-Orthogonal-to-Skin algorithm
5. Filtering: Bandpass filter to isolate heart rate frequencies
6. Estimation: FFT analysis to find dominant frequency (heart rate)

• Ensure face is clearly visible and frontal
• Check lighting conditions
• Try the face detection test first
• Consider different camera angles

• Reduce confidence threshold
• Increase window size
• Check for excessive movement
• Ensure adequate video length

• Improve lighting conditions
• Reduce head/camera movement
• Use longer analysis windows
• Check face visibility throughout video

• app.py - Main Gradio application with rPPG implementation
• requirements.txt - Python dependencies
• README.md - This documentation
• Archive files (app_old.py, etc.) - Previous VitalLens-based implementations

• Reliable face detection - No mysterious "No faces found" errors
• Self-contained - No external library dependencies with compatibility issues
• Transparent processing - Clear understanding of each processing step
• Better error handling - Informative feedback when processing fails
• Customizable - Easy to modify algorithm parameters and methods
• Efficient processing - Optimized chunk-based algorithm for faster results