Remote Multi-Object Detection & Tracking for Retail Surveillance System
Worked on a large-scale computer vision dataset for a retail surveillance AI system designed to monitor in-store customer activity and inventory movement. Annotated over 120,000+ video frames and 45,000+ images using bounding boxes and multi-object tracking techniques. Labeled multiple object classes including persons, shopping carts, products, and restricted-area entries. Performed frame-by-frame tracking to maintain object identity consistency across sequences. Ensured accurate labeling for occlusions, motion blur, and low-light conditions. Collaborated with ML engineers to prepare datasets for training YOLO-based object detection models, improving detection accuracy and reducing false positives. Quality Control Measures: Maintained 98%+ annotation accuracy Followed strict labeling guidelines and class taxonomy Conducted peer-review validation Performed consistency checks across video sequences Verified bounding box tightness and correct object classification