Mobile Robot Based On Monocular Vision

Because of the advantage of unique simple realization, fast perceptual speed and high robustness for visual sensor, the mobile robot is researched and explored by scholars at home and abroad. However, because of the complexity of the visual sensors image information and the instability factors of system, mobile robot localization based on visual system still exists bigger defects and difficulties. In order to acquire higher accuracy rate, this paper mainly carry on research about monocular vision of positioning of the mobile robot in unknown environment. The main research content is as follows:First of all, the paper briefly describes the research background and significance of the subject at home and abroad, through the analysis of theoretical knowledge and the research status, put forward the existing technical problems of vision mobile robot localization algorithm, at the meantime, Build relevant model based on the mobile robot vision system.Second, SURF algorithm gains extensive attention after it is put forward and used to image local feature extraction. The quantity of the feature points extracted by SURF algorithm is large, with the larger perspective and noise changes in image, extraction of feature points by unstable defects is improved, and the Hermite matrix is proposed based on improved discrete Gaussian- SURF image matching algorithm. Through the Dual-tree Complex Wavelet Transform, low frequency part of the image is the input of the improved SURF algorithm. Calculating the image Gaussian-Hermite moment and description vector of interest points, besides, using 3D non-maxima suppression to determine feature points at different scales. Through MATLAB simulation, the experimental performance of the algorithm is verified. The experimental results show that the improved algorithm decreases the number of feature points, and the accuracy of matching is improved.Again, in view of the poor real-time performance and the low correct matching rate of the image feature points. Using the dot product vector to replace the Euclidean distance measurement method to measure the degree of similarity between descriptors is proposed; In order to further improve the search efficiency and accuracy, random KD tree algorithm of feature point matching by the PROSAC algorithms eliminate false match is used. According to the simulation experiments on the improved algorithm performance analysis, the validity of the algorithm and real-time performance is proved.Finally, the improved algorithm is combined with the adaptive particle filter algorithm; the mobile robot vision orientation in indoor environment is realized. In view of the large amount of calculation particle filter algorithm, the particle impoverishment of defects, the figure of the similarity coefficient is used to update the adaptive filter algorithm in the weight coefficient of each particle. Through to adjusting the particle number timely, the amount of calculation is reduced and the performance of system real-time is improved.