yolov8_ros

ROS wrapper for yolov8_tracking

This is a ROS compatible version of the yolov8_tracking repository, which provides the option to use a host of tracking algorithms on the output of yolo models for robust and real-time detection, segmentation and tracking of upto 20 classes of objects. The information of the resulting bounding boxes is published over a ROS node to be used by the controller to navigate the robot using visual servoing.

Usage

1. Clone this repo in your catkin_ws/src
2. cd yolov8_ros
3. Make a copy of the track.py file in the folder yolov8_tracking and save it wherever you like.
4. Now delete the yolov8_tracking folder and clone a fresh copy from yolov8_tracking by following the instructions there. This step downloads the yolov8 repo as submodule
5. Now replace the new track.py file with the one you copied
6. catkin build or catkin_make in the folder catkin_ws/src
7. Now launch the my_tracker.py node (in the src folder) and the track.py script by specifying whatever source, classes etc. (for this refer yolov8_tracking )

Further

Though the monocular framework (fused with 2D Lidar stream of a simple rplidar) gives pretty good results, this can be further improved. Depth data from an intel realsense or ZED stereo-camera can greatly improve performance and obstacle avoidance capabilities. Also currently, the controller latches onto the first person to enter the frame (after a specific time of no detections), however one can choose which person(or object) to follow by simply providing the ID number of the desired bounding box, or writing some script to use natural language commands to decide and pick an ID (eg. "Follow the dog or the guy in the red shirt"), or even use some pre trained identification model to always follow a specific "owner(s)".

Person_Follow.mp4