How to spot hazard debris on a racing track?
100% of the sport depend on track clearance and spectator placed in safe areas.
Obvious hazardous objects or spectators localization (open roads) are reported by marshalls and drivers.
Danger comes when unexpected event happen out of line of sight and beyond human senses.
Machine -car, moto- perception is not used to detect unspotted objects or to predict its drop point if it goes off the track.
To solve this computer vision problem, Yolov5 model created by Ultralytics has been adapted. Indeed, Yolov5 is natively not dedicated to any “content” from motorsport. Thus, a racing Yolov5 model has to be trained on a customized racing dataset. The latter is a synthetic dataset.
Development made with Python. Computation made on AWS instance.