Automated Object Counting in Crowded Scenes

Developed a computer vision application that counts unique people in videos or live webcam feeds using YOLO (You Only

Look Once) for person detection and CSRT (Channel and Spatial Reliability Tracker) for object tracking. Key features include:

● YOLO-based Detection: Utilized Ultralytics' YOLO model to detect people with adjustable confidence thresholds.

● Tracking Logic: Implemented a custom tracker to follow individuals across frames using Intersection-over-Union

(IoU) for matching detections.

● Streamlit Web Interface: Built an interactive dashboard for real-time visualization, parameter tuning, and data logging.

● CLI Support: Added a command-line tool for batch processing videos with configurable thresholds.

● Performance Metrics: Tracked and displayed unique counts over time with a live counter and frame-by-frame analytics.