human_tracking_3D

human detection and tracking in 3D space based on Yolov3 and deepsort

requirements

• ros-kinetic
• usb camera or intel realsense d435i camera
• python2.7

installation

• download darknet_ros package

  mkdir -p ~/catkin_ws/src
  cd ~/catkin_ws/src
  git clone --recursive https://github.com/leggedrobotics/darknet_ros.git
  cd ../
  catkin build darknet_ros

• download deepsort-yolo package and rviz visualization package

  cd ~
  git clone git@github.com:EcustBoy/human_tracking_3D.git
  cp -r ./human_tracking_3D/* ~/catkin_ws/src
  cd ~/catkin_ws/src
  catkin build sort_track
  cd ../
  catkin_make

run

• launch camera device

for usb camera

roslaunch usb_cam usb_cam-test.launch

for d435i camera

roslaunch realsense2_camera rs_aligned_depth.launch

• launch darknet_ros

  roslaunch darknet_ros yolo_v3-tiny.launch

  make sure the camera publish topic name is same as darknet_ros subscribe topic name

• launch deepsort node

  roslaunch sort_track sort_deep.launch

• calculate human position in 3D space (core)

  cd ~/catkin_ws/src/sort-deepsort-yolov3-ROS/sort_track/src
  pyhton2 ./my_track_deep.py

  launch node which subscribes yolo detection results and calculates human 3D position coordinate based on camera equation publish (x,y,depth,id) for each human

• visualize in rviz

  rosrun visualize_bbox visualize_bbox

  you can see the real-time human detection results(bounding box) in rviz