Crowd Counting from Shaky Handheld Video - YOLO / CV / DeepSort

The client needed a system to accurately count people in a crowd from a handheld, shaky video. This included: calculating the total number of people and tracking how many crossed a fixed vertical line in the scene.

The project involved developing and implementing a robust multi-stage pipeline that integrated video stabilization, detection, and tracking techniques. Special attention was given to compensating for camera movement and ignoring distractions such as moving flags, ensuring precise crowd counts and accurate line crossing measurements.

The video attached is only the first 30 seconds.