Research On Image Processing And Analyzing For Autonomous Navigation Of Mobile Robot In Outdoor Environments

The improvement of the vision-based navigation ability of the mobile robot is ofgreat significance in the promotion of the mobile robot in the unknown environment.This dissertation designs a vision-based navigation frame, designs and realizes atraversable region recognition system through which traversablitycan be recognized bythe mobile robot quickly and accurately, and thus lays the foundation for autonomousnavigationofmobilerobotinoutdoorenvironment.Firstly, the dissertation introduces the key techniques in autonomous navigationsystems of mobile robots, such as Visual Odometry, Terrain Recognition, ImageSegmentation, Traversable region Recognition. Also, the dissertation proposes avision-basednavigationframebasedonthesetechniques.Secondly, an improved fast outdoor color image segmentation algorithm based onmean-shift is presented, and the optimum parameters of the image segmentationalgorithm are found through the experiments and theory analysis. The image scaletransformationisusedtoincreasetheefficiencyofthealgorithm,becausethismethodisapplied broadly in the image segmentation area, and especially adapts to the systemwhich requests more celeritythan accuracy. The experiments carried out on the outdoorcolor images show that good segmentation results can be achieved by the proposedalgorithm.Finally, the dissertation studies the traversable region recognition in outdoorenvironment. A traversable region recognition algorithm is presented combining theSVM method widely used in pattern recognition with outdoor environment imagesegmentation presented in this article. The images used in these experiments arecollected in the school environment. The main parameters of feature extracting andSVM training are optimized. According to the experiments, the probability whichsucceed in classing the testing images reached 85%, and the traversability can beobtainedaccurately.