Tracking Loss for Small Targets (Boxes) with Irregular Motion and Occlusion using YOLOv8 + ByteTrack

[dixingxing1]

I'm working on a project where I need to track multiple boxes, each with a unique code, as they are moved by people in a warehouse area. I'm using YOLOv8 for detection and ByteTrack for tracking.

However, I'm facing an issue where the tracker easily loses targets, especially during sudden turns or when occlusions occur (e.g., when a person blocks the view of a box while pushing it). I suspect this is due to a combination of factors:
Small target size:​ The boxes can appear relatively small in the video frames.
Irregular motion:​ The movement path of the boxes is not linear, especially during sharp turns pushed by people.
Occlusions:​ Targets are frequently blocked by people or other boxes.
Has anyone encountered similar challenges or have suggestions for improving tracking robustness in such scenarios? Any advice on parameter tuning for ByteTrack, modifications to the tracking logic, or alternative approaches within the YOLOv8+ByteTrack framework would be greatly appreciated.

Please let me know if you need any further details about my setup. Thank you!

[LamnouarMohamed]

i have a suggestion for you, you can add to your algorithm appearance model.

[dixingxing1]

i have a suggestion for you, you can add to your algorithm appearance model.

I have tried to enhance the yolo detection, but still lose track when a worker pushes a box around a corner with their body hiding the box now and then. The boxes look alike. Any suggestion? Thank you in advance.

[LamnouarMohamed]

i didn't understand can you re-explain or give some images about your situation

[dixingxing1]

i didn't understand can you re-explain or give some images about your situation

We are working on tracking boxes in a factory warehouse. The boxes contain production parts, and each box has its own unique information. We have set up cameras above the factory floor for tracking. Workers move the boxes within the warehouse, resulting in irregular movement, and occlusions occur.