CCTV-Vision

Video Person Search System - Search hours of footage by appearance instead of scrubbing frame-by-frame. Advanced YOLO-v8 detection with custom trained EfficientNet-B4 classification for gender, shirt color, hair color using Optuna optimization.

Person Detection & Tracking Attribute Classification Query Engine

Additional Demonstrations

Individual person detection with clothing color analysis focusing on shirt color classification and gender identification for targeted person search queries.

Single Person Attribute Classification

Gender classification system using EfficientNet-B4 model to identify male/female attributes along with hair and clothing color analysis for comprehensive person search.

Gender & Attribute Classification

High-motion sports tracking demonstrating system performance during rapid movement with clothing color detection and persistent ID tracking in dynamic environments.

Dynamic Movement Tracking

Color-based person search demonstration using Lab color space and ΔE2000 distance calculation to accurately identify distinctive clothing colors like red shirts.

Color-Based Person Search

Extended duration tracking showing system's ability to maintain persistent IDs and analyze athlete attributes over longer sequences with MediaPipe pose detection.

Extended Duration Analysis

Real-World Performance

95.25%
Detection Accuracy
93.20%
Attribute Classification
8-15
FPS Processing
ΔE2000
Color Distance
EfficientNet-B4
Neural Backbone
PA-100K & PETA
Training Datasets
Optuna
Hyperparameter Tuning
+22pp
Above Baseline