ros detection and segmentation module

This is a ROS node for object detection and segmentation from 2D image to 3D point cloud. The detection is based on YOLO-v2. The segmentation is based on SegmenterLight.

Environment

The RGBD camera is Astra of ORBBEC and the ROS SDK is supposed to be installed first.

Compile and prepare

YOLO-darknet

Note that more details about the instructions and the code of ROS-YOLO can be found here. Sorry for the confusion.

Download the the pretrained model from YOLO-v2 under the path ./darknet and ./data.

Change the path on line 231 in the file image.c in ./darknet/src to your own path. Replace all XXX/catkin_ws path in the file ros_yolo_test.cpp and test_detector.h with your own ROS workspace path.

Compile the darknet module.

• cd darknet
• make

SegmenterLight

Run the following commands to compile and install. Change the topic name to registered_cloud on line 244 in pcl_create.cpp.

• cd SegmenterLight
• mkdir build
• cd build
• cmake ..
• make
• sudo make install

Compile the ros_yolo node and ros_seg node: catkin_make.

How to run

• roscore

Run Astra ROS SDK.

• roslaunch astra_launch astra.launch

Run the YOLO node.

• rosrun ros_yolo ros_yolo_test

Check the published YOLO message.

• rostopic echo /yolo_ret

Register the point cloud to image and create the segmented object point cloud.

• rosrun ros_seg registered
• rosrun ros_seg pcl_create