Traffic Accident Detection (Computer Vision)

• Challenge: Limited access to critical traffic camera data necessary for accident detection research.

• Approach: Curated a comprehensive dataset by sourcing publicly available footage and annotating accident

scenes. Utilized models such as YOLOv5, Faster R-CNN, and SSD_mobilenet_v2 to train the system.

• Outcome: Chose YOLOv5 for its superior accuracy (92%), optimizing it for real-time detection with minimal

false positives.