The Research On Mobile Robot Localization Methods Based On Stereo Vision

The development of autonomous navigation technology makes mobile robot has greaterflexibility and a broader application prospects, and therefore the study of this technology hasa very important significance. In order to achieve autonomous navigation of mobile robot inunknown environment, the robot must have the ability of localization and composition.Localization is a prerequisite for the completion of robot navigation. Conventionallocalization methods typically include the global positioning system (GPS) and the deadreckoning. The former can not be used in case of GPS failure, and the latter is difficult toresolve the problem of the encoder reading error which is caused by the wheel slip. Based onthe above background, this paper carried out the research on mobile robot localizationmethods based on stereo vision. Multiple binocular image sequences, which were captured bythe camera, were calculated to draw robot localization information. This method makes up forthe deficiencies of traditional methods and lay the foundation for autonomous navigation.To solve the self-localization problem, we mainly do the following work: binocularcamera calibration, images feature extraction and stereo matching, images feature tracking,motion estimation and so on.In this subject, the requirements of positioning accuracy and real-time performance needto be met. Harris operator was used for image feature extraction because of its relatively fastoperation speed. Then, normalized cross-correlation matching was used to preliminary matchthe images feature points to remove the feature points on the large differences. After that,RANSAC algorithm was proposed to eliminate the erroneously matching points. Furthermore,on the basis of Zhang Zhengyou calibration method, we calibrated the camera in the imageprocessing software Halcon10.0, this method with fast computing speed, less susceptible tointerference and high accuracy, can mainly meet the demands of robot localization. Pixelcoordinates of matching points were compared to achieve feature tracking, then thosecoordinates were used to calculate the3D coordinates of the feature points of the front andrear frame, and high precision tracking results obtained by this method. In the motionestimation part, the least squares method, which was solved by singular value decomposition,was used to estimate the motion parameters. Then the estimates were used as the iterativeinitial values of the nonlinear least squares estimation to calculate accurate estimation results.Experiments Based on Matlab show that the algorithm has good real-time characteristicand higher accuracy. It is beneficial for the follow-up study.