Obstacle Detection Of Intelligent Wheelchairs Based On Machine Vision

With the aging of society and the increasing caring for disabled people, robots serving the elderly and the disabled are becoming a focus of research. Among such service robots, intelligent wheelchair shave attracted great attention both home and abroad for integrating cutting-edge technologies in the fields of robots, IT, pattern recognition and image processing. Meanwhile, intelligent wheelchair security has become a paramount issue. Its top priority is the timely detection of obstacles and pedestrians on the route of wheelchairs, which is a hot topic in intelligent wheelchair research.Machine vision-based obstacle detection is one of the advanced, fast-developing technologies. Compared with non-visual censors, the vision-based detection boasts wider detection range and provides more needed information. This thesis, taking intelligent wheelchairs as its object of study, tries to explore key technologies to the detection of obstacles and pedestrians with the help of binocular stereo vision technology and pedestrian detection technology. The main research steps are as follows:The thesis first introduces some key technologies in obstacle detection of intelligent wheelchairs, briefing on the research status of camera calibration, binocular matching and pedestrian detection in China and abroad. It then elaborates the fundamental principles of binocular stereo vision ranging on the basis of the two optical axis models of binocular stereo vision and after a comparison, selects the parallel binocular optical axis model which is easier to compute.Moreover, it contrasts and analyzes common camera calibration algorithms and proposes the checkerboard sample screening method. This method makes use of relations among corners of a checkerboard to exclude wrong corner detection samples, thus increasing the accuracy of calibration results and at the same time getting the intrinsic and extrinsic parameters of a binocular camera.The thesis points out the respective strengths and weaknesses of the two major pedestrian detection methods and studies the pedestrian detection that is based on histogram of oriented gradients and support vector machines classifiers. By means of such detection, the position of a pedestrian is revealed through detecting the image sequences acquired by a camera.In addition, the thesis compares several stereo matching methods and adopts the matching algorithm based on dynamic programming, and attains the disparity map by using camera calibration results. It then calculates the distance by combining camera imaging principle,3D ranging principle and triangulation measuring principle.At last, it conducts a simulation of obstacle detection on the basis of vs2012 and OpenCV2.4.6. The simulation proves that this method applies to settings requiring the detection of various types of distances and handles pedestrian detection in a special manner. All this paves the way for a better control of intelligent wheelchairs.This paper studies methods of obstacle detection of intelligent wheelchairs from the perspectives of binocular stereo vision technology and pedestrian detection technology. Its object of study being intelligent wheelchairs, the findings and methods of this research are also valuable to other intelligent mobile robots in terms of their navigation and ranging modules.