Sharpiless/Yolov5-deepsort-inference

Combines YOLOv5 object detection with DeepSort's re-identification features to maintain consistent track IDs across frames, enabling accurate counting of specific object classes (person, car, truck). The implementation uses half-precision inference and non-maximum suppression for real-time performance, while DeepSort's appearance-based matching with configurable distance thresholds handles occlusions and identity switches. Supports custom model training and weights integration, allowing users to adapt the pipeline to domain-specific detection tasks beyond the pre-configured vehicle and pedestrian classes.

1,459 stars. Actively maintained with 1 commit in the last 30 days.