Backend Python Scripts Documentation

Module: `object_detection.py`

Purpose: YOLO-based object detection

Key Functions	Description
`predict_from_cv2_frame()`	Performs object detection on a frame using Ultralytics YOLO model

Key Features	Details
Image Conversion	Converts OpenCV frames to PIL images
Image Resizing	Resizes images to 640x640 for YOLO compatibility
Detection Output	Returns filtered bounding boxes with class names and confidences

Module: `threshold_config.py`

Purpose: Manage and persist threshold configuration

Class: `ThresholdConfig`

Configuration Factors	Description
`blur_factor`	Sensitivity to image blur
`movement_factor`	Sensitivity to object movement
`shake_factor`	Sensitivity to camera shake

Key Methods	Functionality
`load_config()`	Load configuration from JSON
`save_config()`	Save configuration to JSON
`update_factors()`	Update individual factors

Module: `video_analyzer.py`

Purpose: Analyze video characteristics for detection

Class: `VideoAnalyzer`

Key Analysis Methods	Description
`analyze_video()`	Processes entire video to determine baseline metrics
`_calculate_thresholds()`	Computes detection thresholds

Metric Calculations	Details
Blur Detection	Laplacian variance
Movement Tracking	Movement area calculation
Shake Metric	Shake computation
Threshold Calculation	Outlier removal for robust results

Module: `video_processor.py`

Purpose: Process videos for machine learning dataset creation

Key Functions	Description
`process_video()`	Main video processing function Extracts frames based on movement and quality Generates YOLO-compatible annotations Interacts with S3 for dataset management
`split_data()`	Splits processed data into train/test/validation sets

Advanced Features	Details
Class Management	Dynamic class name management in YAML
Data Split	Configurable train/test/validation split ratios
Storage	S3 integration for dataset storage
Frame Selection	Based on movement, blur, and shake thresholds