Object detection

[mrc-210]

Hi everyone,
i'm trying to using YOLOV8 for objects tracking, but i have any problems.
The goal of my project is to identify objects (static and dynamic) and keep track of their center of mass.

1. YOLOV8 often associates more than one building box with the same object. Since I use the pre-trained model for my experiment, I wondered if training the model on an experiment-specific dataset could solve this problem. If not, how could I solve it?
2. When tracking a dynamic object, it can happen that the object gets occluded and consequently the size of the associated building box changes. Since I use building boxes to define the center of mass of each object, I was wondering if there was a way to avoid this problem.

Thank you for the answers

[glenn-jocher]

@mrc-210 hi,

1. Training the model on a dataset specific to your experiment can indeed improve detection accuracy and reduce duplicate bounding boxes. Additionally, fine-tuning the Non-Maximum Suppression (NMS) settings may help.
2. To handle occlusions and maintain consistent tracking, consider using a tracking algorithm like ByteTrack or BoT-SORT, which are designed to handle such scenarios effectively.

For more details, you can refer to our documentation on object tracking: https://docs.ultralytics.com/guides/object-counting/

Best of luck with your project!

[mrc-210]

@glenn-jocher i'll try, thanks you so much!

[glenn-jocher]

@mrc-210 you're welcome! If you have any more questions, feel free to ask.