Research And Application Of SLAM Method Based On Mobile Vision Sensor

With the progress of image processing technology,the research on simultaneous localization and map building(SLAM)has gradually become a hot topic in the field of computer vision.Nowadays,the performance of mobile devices is constantly improving,which makes SLAM research based on mobile vision sensor possible.Mobile devices are flexible,powerful,and easy to develop.SLAM system development using mobile devices is of great practical significance.This paper mainly studies the SLAM method of monocular camera commonly used in mobile terminal and the application development of SLAM system based on iOS platform.In the research of algorithm,based on the classical visual SLAM method,this paper focuses on the visual SLAM method based on feature point method.ORB algorithm is used in feature extraction.The improved FAST algorithm is used to extract key points,and the improved BRIFT algorithm is used to extract binary descriptors,which speeds up feature extraction and reduces memory requirements for storing features.The fast approximate nearest neighbor(FLANN)feature matching algorithm is improved in feature matching.This algorithm uses the improved K-means algorithm to cluster features,and builds multiple hierarchical clustering trees to search for features,which improves the accuracy and efficiency of feature matching.The comparison experiments show that the improved FLANN algorithm has good invariance and robustness.The random sampling consistency(RANSAC)algorithm is improved in mismatching elimination.This algorithm improves the problems of low efficiency,easy degradation and poor accuracy of classical RANSAC algorithm,and reduces the probability of mismatching.Through comparative experiments,it is verified that the improved RANSAC algorithm can improve the accuracy and speed of feature matching after mismatching elimination in different scenarios.At the same time,the improved feature matching and mismatching elimination algorithm is applied to loop closing detection,which improves the effect of loop judgment through key frame feature matching.In application,this paper designs and implements a SLAM system based on iOS platform.The system is designed with four modules: initialization,visual odometry image processing,loop closing detection,and back end optimization.In the development process,the three threads of tracking,trajectory and optimization are used to complete the development of the above modules.By running the system,it can be seen that the mobile test platform developed by using the improved algorithm can draw the trajectory in real time,correctly judge the loop closing,and ensure that the generated trajectories are globally consistent.