pcl_recognition

A package for object recognition, build for EECS 476 Mobile Robotics Individual Project

Collaborators

Ran Hao ([email protected])

Shipei Tian([email protected])

Things to do before running code

Plug in your Kinect to your machine

Start Kinect driver: roslaunch freenect_launch freenect.launch

Open rqt_reconfigure: rosrun rqt_reconfigure rqt_reconfigure

In camera / driver, check the box depth_registration.

Object Recognition Kitchen

Since now, the most fast and accurate way to recognize object is using ORK which is a mesh based recognition, see (http://wg-perception.github.io/object_recognition_core/).

####Run it by rosrun object_recognition_core object_add.py -n "coke " -d "A empty coke can" --commit

rosrun object_recognition_core mesh_add.py [the object id that previous command returned] `rospack find pcl_recognition`/data/coke.stl --commit

roslaunch pcl_recognition ORK.launch

Video demonstration: https://youtu.be/lbvA26GHWz0

PCL Approaches

This packages contains multi approaches from PCL to recognize objects in a Kinect scene, including:

Correspondence Grouping

(http://pointclouds.org/documentation/tutorials/correspondence_grouping.php)

####Run it by

roscd pcl_recognition/pcd

For test 1:

rosrun pcl_recognition correspondence_grouping coke_bad.pcd coke_scene.pcd -k -c --model_ss 0.01 --scene_ss 0.01 --cg_thresh 2.9 --cg_size 0.09

For test 2:

rosrun pcl_recognition correspondence_grouping new_coke.pcd coke_2.pcd -k -c --model_ss 0.02 --scene_ss 0.02 --cg_thresh 2 --cg_size 0.13

Implicit Shape Model

(http://pointclouds.org/documentation/tutorials/implicit_shape_model.php)

####Run it by

roscd pcl_recognition

./ism_command.sh

Hypothesis Verification

(http://pointclouds.org/documentation/tutorials/global_hypothesis_verification.php)

####Run it by

roscd pcl_recognition/pcd

rosrun pcl_recognition global_hypothesis_verification milk.pcd milk_cartoon_all_small_clorox.pcd -k

Iterative Closest Point

(http://pointclouds.org/documentation/tutorials/interactive_icp.php)

roscd pcl_recognition/pcd

rosrun pcl_recognition icp coke.ply 20

Library for recognition

From all the PCL Algorithms, the best one is Correspondence Grouping using Hough, a library is build with that algorithm, see object_recognizer.cpp, for usage see object_recognize_main.cpp

####Run it by

For test 1:

roslaunch pcl_recognition object_recognizer_test1.launch

For test 2:

roslaunch pcl_recognition object_recognizer_test2.launch

For test 3:

rosrun pcl_recognition object_recognize_main

PCD Edit Tool

For all approaches in PCL library, it all requires pcd files as original input, so we write an useful tool to make a pcd file from Kinect cloud so that you can use it as input for object_recognizer.

####Run it by

rosrun pcl_recognition pcd_edit_tool name_you_want_to_save.pcd

Usage demonstration: https://youtu.be/GV69MXoV2kg

Calculate normal, centroid of a plane

####Run it by

roslaunch pcl_recognition find_stool_coke.launch

Select the object you want to find using Publish Selected Points.

Or, you can this node separately by:

rosrun pcl_recognition find_stool (your pcd file name)

It will let you select the type of object you want to find in order to load the best filter range, if other selection, it will let you manually input the filter range