Research On Robot Monocular Vision Localization Based On Improved Bow Model

Nowadays,more and more robots are applied to people's life fields.With the rapid development of artificial intelligence and image processing technology,robots are more in telligent.SLAM(simultaneous localization and mapping)is the basic and key problem of autonomous action of robot.The image obtained by visual sensor contains rich environm ental information.Therefore,the implementation of SLAM task through visual informatio n(Visual Simultaneous Localization and Mapping,VSLAM)has gradually become a rese arch heat in this field.Point.According to the number of cameras contained in the syste m,VSLAM is divided into monocular vision and multi vision.In this paper,the monocu lar vision SLAM system based on graph optimization is studied.On this basis,the featur e extraction algorithm and the closed loop detection module are improved to improve the positioning accuracy of the system.First,the SURF,Harris feature extraction algorithm and common feature point descri ptors are studied,and their advantages and disadvantages are summarized,and the combi nation of SURF and Harris is proposed to extract the image feature points.Gauss Pyram id Pyramid is established to obtain the image,and the key points are detected on variou s scales by using the Harris algorithm.The key points are combined with the key points obtained by the SURF algorithm.Then the r BRIEF algorithm is used to generate the des criptors of the key points.Finally,the two way nearest neighbor nearest neighbor ratio method and the RANSAC double filter algorithm are used to remove the characteristics.Mismatch between points to obtain reliable and stable feature points.Then the basic principle of closed loop detection in SLAM is studied,and the grow th of dynamic word bag tree is completed by using the key frames obtained in the proc ess of robot movement to improve the performance of the closed loop detection module.On this basis,the similarity between the traditional similarity scores is calculated only on a single scale and the accuracy of the image matching is low.The Pyramid matching al gorithm is introduced to evaluate the similarity between the images at different levels by using the TF-IDF scoring standard,so as to improve the accuracy of the similarity evalu ation between the images,according to the similarity.The candidate closed loop is obtai ned by the sex score,and the false positive loop is removed by RANSAC.The data set of Visual Geometry Group,University of Oxford,City Center and Ne w College dataset are used to validate the improved algorithm proposed in this paper.KITTI data sets are used to verify the positioning accuracy of the whole SLAM as part of the improved part of this paper.The results show that the algorithm can improve the a ccuracy and robustness of robot localization.