Research Based On Vision Outdoor Localization Algorithm For Mobile Robot

Because of the complexity of the outdoor visual sensors image information and the cumulative error of the system itself,mobile robot localization based on visual system still exists difficulties.In this thesis,researchers were conducted in the known outdoor environment on the localization system mentioned above to improve the efficiency and accuracy.The main content is as follows:First of all,relevant research progresses at home and abroad were introduced as well as the significant of the subject.Through the analysis of the theoretical knowledge and the research status,key technical problems in the mobile robot localization algorithm based on visual system were expatiated,and an ideal model for the outdoor localization of the mobile robot was established.Secondly,in view of the large number of feature points extracted by SIFT algorithm and the poor stability for the key points of the main direction when the image has scale and rotation changes,the number of the pixels that matched with the feature points was reduced and the first order central moment was used to determine the main direction of the key points in the phase of feature points detection,and then the efficiency of the algorithm is improved.The performance of the improved algorithm was verified by the experiential results.Thirdly,in view of the high eigenvector and low correct matching rate of the image feature points,the sparse feature representation method was introduced in the process of describing the vector by the key points to extract the eigenvector of SIFT key point,which the high dimensional gradient derivative vectors were transformed into low dimensional sparse feature vectors.In order to further improve the accuracy of the algorithm,the disparity constraint algorithm was used to eliminate the false matching pairs during the feature points matching process.In the end,the performance of the improved algorithm was analyzed and compared with the simulation,and the accuracy and real-time performance of the improved algorithm are verified.Finally,the mobile robot localization based on visual system in the known outdoor environment was realized by the combination of EKF and UKF particle filter algorithm.Regarding to the particle impoverishment of defects,the state information of fusion particle filter was proposed to get better particle distribution.The similarity coefficient of the feature matching was used to update the weight coefficient of the particle filter,which reduced the computational complexity of the algorithm and the performance of system's real-time is improved.