Determining the Viability of the Kinect as a Research Tool

[2011] The Defense Presentation for my MAS R&D1

Transcript

1. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Matthew S Roscoe Dr. Paul G Plöger Dr. Ken Kent Determining the Viability of the Kinect as a Research Tool
2. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Goals: •  Determine the usability of the Kinect in Mobile Service Robotics. •  Test through the application of Environment Segmentation. •  Looking for planar surfaces within environments. •  Operate within a real-time constraint of 30Hz.
3. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Hardware
4. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg The Kinect •  320x240 IR @ 30Hz •  640x480 RGB @ 30Hz •  640x480 RGBD @ 30Hz •  1280x1024 RGB @ 10Hz •  1280x1024 RGBD @ 10Hz •  57° Horizontal View •  43° Vertical View •  ±28° Vertical Axis Movement •  Microphone Array •  Speaker System
5. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Getting Distance Values
6. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Structured Light
7. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg What is RGBD? •  Color Image combined with depth values. •  Typical output from the Kinect. •  May also combine depth with an IR image for use in low light conditions.
8. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Kinect Error Ranges
9. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg •  Provides Data Structures to deal with Point Clouds •  Allows use of CUDA GPGPU b a s e d a l g o r i t h m s f o r processing
10. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg RANSAC •  RANdom Sample And Consensus •  Used to fit models of data sets •  Functions well in extremely noisy data sets
11. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg 1 10 100 1000 1 2 3 4 5 6 7 8 9 10 Running Time (ms) Planes Detected CPU vs GPU End to End Computations CPU END-To-END GPU END-To-END
12. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg REAL WORLD USAGE
13. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg
14. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg
15. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Sunlight
16. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Outdoor Use
17. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Shadows
18. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Conclusions •  Useful within a very specific range. •  If you are looking for long range values laser scanners are best. •  Not advisable for path planning. •  Not usable outdoors. •  Multiple Kinects have interference issues. •  Has issues dealing with varying types of materials.
19. Fachhochschule Bonn-Rhein-Sieg Fachhochschule Bonn-Rhein-Sieg Conclusions •  A cost effective way to gather distance, color, and sound information from 0.8m until 2.5m