0. Prerequisites

    Ubuntu 
    CUDA
    OpenCV
    gstreamer 
    ROS (gscam)
   

1. Clone or copy darknet_det code into a ROS worksapce. Suppose the directory tree is as below:

        [ros_ws]/src/darknet_det/ 
   

   from: https://github.com/maoxuli/darknet_det

2. Clone or copy darknet code to some directory. For example, inside darknet_det directory, or side by side, as shown below:

        [ros_ws]/src/darknet_det/darknet/ 
   
    or  [ros_ws]/src/darknet/ 
   
    or  in other directory 
   

   from: https://github.com/maoxuli/darknet

    or  https://github.com/pjreddie/darknet  
   

3. Build darknet

    cd [ros_ws]/src/darknet_det/darknet     (according to above step)
   
    edit [ros_ws]/src/darknet_det/darknet/Makefile as below (or according to your situation): 
   
        GPU=1
        CUDNN=1
        OPENCV=1
        OPENMP=1
        DEBUG=0
   
    make 
   

4. Build darknet_det

    set darknet path in [ros_ws]/src/darknet_det/CMakeLists.txt, according to above steps: 
   
        set(DARKNET_PATH ${CMAKE_CURRENT_SOURCE_DIR}/darknet)
   
    cd [ros_ws] 
   
    catkin_make 
   

5. Download pre-trained YOLO networks (cfg file, weights file, and names file) to directory shown as below:

        [ros_ws]/src/darket_det/models/yolov3.cfg 
        [ros_ws]/src/darket_det/models/yolov3.weights 
        [ros_ws]/src/darket_det/models/coco.names 
   

   from: https://pjreddie.com/darknet/yolo

    edit darknet detection settings accordingly in [ros_ws]/src/darknet_det/launch/darknet_det.launch, e.g.:
   
        <arg name="cfg_file" default="$(find darknet_det)/models/yolov3.cfg" />
        <arg name="weights_file" default="$(find darknet_det)/models/yolov3.weights" />
        <arg name="names_file" default="$(find darknet_det)/models/coco.names" />
   
    set target object classes for detection as below: 
   
        <rosparam>class_ids: [ 0, 3, 14 ]</rosparam>
   
    the number is the class ID defined by dataset used for network training, e.g. in COCO dataset, ID 0 is for person. 
   

6. Test with CSI camera on Jetson Xavier/TX2/Nano or Raspberry PI

    cd [ros_ws]
   
    source devel/setup.bash 
   
    edit camera settings in [ros_ws]/src/darknet_det/launch/csi_cam_det.launch 
   
    edit darknet detection settings in [ros_ws]/src/darknet_det/launch/darknet_det.launch 
   
    rolsuanch darknet_det csi_cam_det.launch 
   

7. Understanding detection result

darknet_det publish deteciton result on topic [camera_name]/det. Please refer to [ros_ws]/src/darknet_det/msg for defined message type. Please note that the bounding box in current implementation is denoted with center point (x, y) and size (w, h), normalized to image size.