Object tracking in image coordinates

Implement a object tracker for later usage in camera readjustment that object is always in image center (see #84 (closed) ).

Task is split to two main tasks

1. Tryout current StateOfTheArt

TODOs check models on a video and check which provides best usage characteristics on our data

• yolo10+botsort/bytetrack : https://docs.ultralytics.com/modes/track/#tracking
• https://ai.meta.com/sam2/
• https://github.com/IDEA-Research/Grounded-SAM-2
• https://docs.ultralytics.com/models/fast-sam
• dinov2 segmentation https://dinov2.metademolab.com/demos?category=segmentation

2. Create Basic Tracker in ROS

• Create YOLO node (to get detections & tracks)
  • Parameters
    • offers all yolo parameters (onnx path, tracking parameters, etc...) from https://docs.ultralytics.com/modes/track/#tracking
    • CPU count to run models (onnx parameter)
    • do_tracking yes/no
  • Inputs
    • Image msg
  • outputs
    • Raw detections as Detection2DArray msg
    • Tracking results as Trackarray msg
      • Fill only vision_msgs/BoundingBox2D box as tracked 2d box
• Add drawing of results in UI, or later in hud_drawing_node
• Add unittest for the node
  • use pytest
  • feed in two images
  • check if one track (or more if more objects) are generated and almost on same pos in second image