• Use KCF/GOTURN/TLD with YOLOv3 to multiple track and data association(not complete), in case of losing targets.
• Use darknet_ros node for YOLOv3 to detect objects, and publish information of result boxes to tracking node.
• Use usb_cam node to drive the camera to detect online, and publish images to darknet node and tracking node.
• Use video_pub node to read the video and publish frames to darknet node and tracking node.

• Ubuntu 16.04
• ros kinetic
• opencv 3.4.1+
• if using yolov3 tracking, you need to install Nvidia driver, CUDA, CUDNN

• First publish images

roslaunch video_pub video_pub.launch

or

roslaunch usb_cam usb_cam-test.launch

• First run darknet_ros node

roslaunch darknet_ros darknet_ros.launch

• Then run tracking node

roslaunch tracking tracking.launch