Race Timing System: Analyze a recording

Analyze a recording

Run the local timing receiver first if you want detections delivered to its review screen. Supply the actual timezone-aware time corresponding to the recording's first displayed frame, not the file modification time. Checkpoint must match one of the local configured checkpoint names.

python3 camera_capture.py --source finish.mp4 --recording-start 2026-10-10T13:00:00Z --camera finish-cam --checkpoint finish --roi 100,100,640,300 --send

ROI is x,y,width,height in source pixels; the example rectangle must be replaced with coordinates appropriate to your video. Use a small, well-lit region around the bib/crossing zone to reduce unrelated text, evidence size and OCR work. Omit --roi to process the full frame, provided the PNG evidence stays under 256 KB.

The default samples up to two source frames per second for up to 600 sampled frames. --fps (up to 30) and --max-frames (up to 10,000) are explicit controls. Higher rates require more decoding/OCR capacity and must be tested. The first qualified bib read in a three-second cooldown is logged; --cooldown changes that interval. Cooldown applies per bib in this invocation, not a race lap rule. Each logged detection carries frame-relative time, absolute time, OCR confidence and PNG evidence. Recording frames use FFmpeg presentation timestamps after normalizing the first frame to zero; they do not use OCR completion time. Validate recording start and frame timing against a known clock before use.