ROS Object Detection and Classification with pmd CamBoard pico flexx

Introduction

This repository contains a ROS node for Object Detection and grasp planning. After naming objects, the program recognizes the objects and claculates a grasping point for the two-finger gripper of KUKA youBot.

See the following video for visualization:

Environment and subscribed topics

For a full environment setup please refer to this document.

The node expects depth image messages under the topic /royale_camera_driver/depth_image and point_cloud2 messages under the topic /royale_camera_driver/point_cloud. These are provided by the official pmd Royale ROS Wrapper.

Published Topic

The node publishes object messages under the topic /object_recognition/recognized_object. The message type is defined in msg/RecognizedObject.msg. An example message is shown below:

header: 
  seq: 1939
  stamp: 
    secs: 1525095329
    nsecs: 744063000
  frame_id: royale_camera_link
name: Duplo
midpoint: 
  x: -0.0169242266566
  y: 0.00806986819953
  z: 0.188075810671
width: 0.033241365099
rotation: -51

Configuration

Configuration is done in parameters/settings.json. The default values in this file correspond to the youBot object recognition pose descibed in parameters/ObjectRecognitionPose.md.

• "debugging": Do you want to see debugging images?

• "objects": Path to object .json fil

• "camera_thresh": Camera distance to ground. (Float in meters)

• "camera_max": Camera distance to highest object. (Float in meters)

• "maximal_contour_difference": Maxiumum difference between contours to be recognized accoridng to Hu-Moments.

• "minimal_contour_length": Miniumum number of a contour's points to be estimated as potential object.

Execution

1. Executing the Learner: roslaunch object_recognition_pico_flexx Learner.launch

2. Executing the Recognizer: roslaunch object_recognition_pico_flexx Recognizer.launch