This package provides sliding-box object detection application using linear subspace method and Circular Color Cubic Higher-order Local Auto-Correlation (C3-HLAC), which was proposed in A. Kanezaki, T. Suzuki, T. Harada & Y. Kuniyoshi. Fast Object Detection for Robots in a Cluttered Indoor Environment Using Integral 3D Feature Table, Proceedings of the 2011 IEEE International Conference on Robotics and Automation (ICRA2011), 2011.
Please also see our project page here.
Get color point cloud
This application works with Kinect. Please install openni_kinect package, first. Connect Kinect and start to get color point could:
$ roslaunch openni_camera openni_node.launch
Three steps for object detection application
1. learn the environment
$ bash `rospack find color_voxel_recognition`/demos/SH/learn_env.sh
Press "Enter" key for saving data, and "Ctrl+C" ONLY ONCE for quit. (This step is necessary for only once.)
2. learn objects
$ bash `rospack find color_voxel_recognition`/demos/SH/learn_obj.sh <object_name>
Put the target object closest from Kinect and confirm the green box contains only the object. Press "Enter" key for saving data, and "Ctrl+C" ONLY ONCE for quit. (This step is necessary once for each object.)
3. detect one object
$ bash `rospack find color_voxel_recognition`/demos/SH/detect_test.sh <object_name> <similarity_threshold> <bounding_box_size>
similarity_threshold should be between 0 and 1. bounding_box_size should be given in metric unit.