10.3 SITL AI yolo obj recog - terrytaylorbonn/auxdrone GitHub Wiki

24.0624

(24.0508 was "Total sim P3a AI Yolo")

The Gdrive document 10.3,4_(1.3ab)_sim-all_ai-yolo_lidar_auton_v48_24.0218.docx (first part "a" ) shows how to implement Yolo3 object recognition.

For the camera in the diagram below:

  • Video 11: Install Darknet_ROS. Nvidia toolkit (I did not). Clone darknet. Build/source. Update ros.yaml (messages). Run sim (roslaunch hills.launch, startsitl.sh, view topics). roslaunch darknet_ros.launch. roslaunch iq_sim apm.launch. takeoff and put person in view (recognized).

    image image

  • Video 12: Write sub.cpp. catkin build. rosrun sub (no manual takeoff; sub.cpp is ROS subscriber). Add table. sub.cpp receives the ROS messages and prints out (red square).

    image image

  • Video 13: Write sr.cpp (search and rescue). Catkin build. roslaunch apm.launch (mavros). roslaunch darknet_ros.launch. move person (copter already in air). Darket node gets camera shots, sub.cpp gets darknet boxes and if person recognized gives land command.

    image image