[Python] Issues with Product Detection and Consistent Person IDs Using YOLOv8 and DeepSORT

I'm working on a real-time object tracking system using YOLOv8 for object detection and DeepSORT for tracking. The goal is to detect both customers (persons) and products, and assign each person a unique and consistent ID even if they temporarily leave the frame and return. Additionally, I want to identify when a customer picks up a product.

Here's the issue I'm facing:

The person IDs keep changing when the person reappears after being temporarily lost by the tracker.

The system is not reliably detecting or identifying the products, even though they are visible in the frame.

I'm using hashlib to generate consistent hashes for bounding boxes, but it doesn’t seem to be enough to preserve identity over time or handle reidentification correctly.

Has anyone encountered similar problems? What would be the best approach to:

Maintain consistent IDs for people, even if they leave and re-enter the frame?

Improve product detection and identification, possibly with a better approach than hashing?

Continue reading...