There are 5 ros package:
/PointCloudDeal
and /ROIviewer
. Objects are detected by simple height threshold./ROIpicture
, but the result is not very good. /ROI2D
is results of the 3D-2Dbox.)The ROS node graph is as below.
The light blue box is result of Lidar. The light red box is result of YOLO. The dark red box is result of fusion.
ROS;OpenCV;PCL
kitti_player/
. You can get from Baidu Wangpan password: w4qk.darknet_ros_old/darknet_ros/yolo_network_config/weights/
. Please do as eggedrobotics/darknet_ros. Or you can also get YOLOv3 weights from Baidu Wangpan password: w4qk.Check the data path and ros topic name (if you use your own dataset).
Finally, catkin_make
.
Punlish kitti data
roslaunch kitti_player kittiplayer_standalone.launch
Point cloud detection
rosrun pcl_deal pointdeal
Three monitoring method are provided.
rosrun pcl_de pclvis
to see point cloud in PCL.rviz
to subscribe /ROIpoint
topic in order to see the 2D grid results.rosrun pcl_de ROIviewer
to monitor the point cloud detection result.3D boxes to 2D
rosrun opencv_deal showROI
YOLOv3
roslaunch darknet_ros_old yolo_v3.launch
Finally, fuse category and location
rosrun depthG depthget
Then, have fun ! You can see the result through /depthMap
topic.
It's a part of my undergraduate thesis. There are likely many spelling mistake and redundant code. And the codes are ugly. But I will be happy if this repo can help you. Please feel free to open an issue if you have any questions.