Real time spatio temporal segmentation of RGBD cloud and applications

• Author / Creator

  Saini, Amritpal S

• There is considerable research work going on segmentation of RGB-D clouds due its applications in tasks like scene understanding, robotics etc. The availability of inexpensive and easy to use RGB-D cameras and computational capabilities of GPUs has lead to development of numerous applications in this area. Recently proposed cloud segmentation methods are either slow in operation or do not operate in an online fashion making them unsuitable for applications in robotics. In this work we deal with the aforementioned problem. We propose a method to perform online segmentation of RGB-D scene. Our framework is built on dense scene mapping methods like Kinect fusion. It allows us to generate accurate and dense depth maps and provide camera pose information. Instead of directly operating on a large 3D point cloud we process individual RGB and depth frames which are assembled in a dense cloud in an incremental fashion. Pose information is used to integrate the segmentation maps into the global label cloud using GPU. We perform multi-view integration of segments as the camera is moved around in the scene by formulating the problem as weighted graph. We will discuss applications of our segmentation framework to perform real time and scalable object discovery and object detection.

• Subjects / Keywords

  • GPU
  • Micorsoft Kinect
  • Object discovery
  • Point cloud
  • RGBD
  • Object detection
  • Segmentation

• Graduation date

  Spring 2015

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/R3TQ1X

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.