Hybrid Vision System For robotic Indoor Navigation

Open Access
Mangalgiri, Anuradha
Graduate Program:
Aerospace Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
December 20, 2012
Committee Members:
  • Lyle Norman Long, Thesis Advisor
  • Stereo Vision
  • Robotics
  • Indoor Navigation
A hybrid vision system was developed for indoor robot navigation. The robot used an all-terrain robot chassis as its base. It was equipped with a minoru stereo camera, a parallax micro controller, sabertooth motor controller and two NimH batteries. A Samsung Q430 laptop was used as the onboard computer. The robot vision system utilized stereo vision for its navigation. Raw images were captured from the minoru stereo camera, in real time and using a feature-based stereo algorithm disparity maps were created. The disparity results were converted into depth results using the camera properties. The depth results were fused with the edge map of the scene. Range values were inferred from this fused image and using the range values in a fuzzy logic algorithm, the intersections were detected. The robot navigation was then based on the detected intersection. To develop the system, various stereo algorithms were explored. Most algorithms provide excellent results for stereo images but fail with raw images from the camera. A feature based algorithm was found to be the most robust for the present vision system. The system was tested in various scenarios. It was successful as a close range system. It was able to detect all the intersections in close range. In the tested scenarios it was able to move avoiding obstacles and walls.