Collision detection system using computer vision on low power devices

A wide selection of stereo matching algorithms have been evaluated for the purpose of creating a collision avoidance module. Varying greatly in the accuracy, a few of the algorithms were fast enough for further use. Two computer vision libraries, OpenCV and MRF, were evaluated for their implementations of various stereo matching algorithms. In addition OpenCV provides a wide variety of functions for creating sophisticated computer vision programs and were evaluated on this basis as well. Two low-power platforms, The Pandaboard and the Beaglebone Black, were evaluated as viable platforms for developing a computer vision module on top. In addition they were compared to an Intel platform as a reference. Based on the results gathered, a fast, but simple, collision detector could be made using the simple block matching algorithm found in OpenCV. A more advanced detector could be built using semi-global stereo matching. These were the only implementations that were fast enough. The other energy minimization algorithms (Graph cuts and belief propagation) did produce good disparity maps, but were too slow for any realistic collision detector.

Autorentext

Andreas Kalvå was born in Stavanger, Norway. He attended the Norwegian University of Science and Technology earning a masters degree in Engineering Cybernetics.In his free time he enjoys cooking and volunteering in his community.