Real-Time Human Tracking and Face Re-identification

– Developing a real-time system for human detection, multi-object tracking, and face re-identification using state-of-the-art

models: YOLOv11 for human detection, BoostTrack++ for smooth multi-person tracking, and AntelopeV2 from

InsightFace for high-accuracy face identification..

– Integrated FAISS (Facebook AI Similarity Search) to index and perform rapid approximate nearest-neighbor searches

over 512-dimensional face embeddings, enabling real-time identity matching at scale.

– Built a dynamic people-counting module to track the number of unique individuals entering predefined zones, enhancing

use cases like attendance and crowd monitoring.

– Fine-tuned the YOLOv11 model on a 16GB custom dataset of human bodies, applying extensive data augmentation to

ensure robustness under various lighting, angles, and occlusions.

– Fine-tuned BoostTrack++’s similarity weights on four custom human body tracking datasets to improve re-identification

accuracy across frames.