There are 5 ros package:
- kitti_player : publish KITTI data.
- pcl_deal : contain
/PointCloudDeal
and/ROIviewer
. Objects are detected by simple height threshold. - opencv_deal : 3D box to 2D. 3D-2Dbox overlap is simply solved by IoU and several rules. (The origin idea is to detection on
/ROIpicture
, but the result is not very good./ROI2D
is results of the 3D-2Dbox.) - darknet_ros_old : Yolov3 detection node. It is a old version of eggedrobotics/darknet_ros. Thanks to their job.
- depthGet : fuse 3D-2Dbox and 2D-YOLObox in picture Coordinate by simple IoU.
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
- Download KITTI dataset into
kitti_player/
. You can get from Baidu Wangpan password: w4qk. - Download the Yolo weight into
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.
-
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.