Research On Fusion Positioning Of Monocular Camera And IMU On Mobile Terminal

Simultaneous Localization and Mapping(SLAM)is a technology that uses sensors to automatically locate and construct maps in real time.Traditional SLAM algorithm is mainly used on PC.The algorithm of SLAM based on vision only is seriously affected by the environment.The algorithm of SLAM based on monocular camera has some problems,such as the rotation cannot be initialized and the scale cannot be determined.The camera and IMU(Inertial Measurement Unit)fusion to solve these problems of monocular camera.In order to make the algorithm more portable,the improved VINS system is transplanted to the mobile terminal on the basis of the best VINS-Mono algorithm which integrates monocular camera and IMU.First in the initialization phase,Harris corner extraction is more time consuming,if applied to the mobile terminal,mobile computing quantity is limited,so first extracted Harris corner extraction Oriented to FAST feature points,and use the pyramid to hierarchical image feature point tracking using LK optical flow algorithm,get the initial position,and by using the visual part of IMU is initialized,get the initial velocity,IMU error uncertainty,as well as the acceleration of gravity.In terms of key frame extraction,the original algorithm is based on parallax and the number of feature points to determine whether to extract key frames.This method produces serious key frame redundancy on the mobile terminal,increases the number of frames to be optimized,and increases unnecessary calculation.In order not to affect the precision,and will not affect the operating efficiency on the mobile end,this paper puts forward improvement strategy of key frame extraction,joined on the key frame extraction conditions in the original algorithm,the overlap ratio calculation is based on feature points overlap ratio calculation is inserted into the key frames,if the number of feature points for two consecutive frames overlap area within a certain threshold,and the middle in the past a certain threshold number of frames in a row,then insert the key frames.The back end continues to use the nonlinear optimization of tight coupling based on sliding window.However,only using the word bag model makes the detection efficiency of return loop low.In this paper,a method of loop detection based on mean similarity is proposed.When the feature points are matched,the mean information is added to improve the efficiency of loop detection and repositioning,and then the global positioning map is optimized.Finally,the real-time positioning system based on mobile terminal is realized.The experimental results show that the initialization speed of the monocular camera and IMU positioning algorithm on the mobile end is 12.5% higher than that of the original algorithm,and the accuracy of loop detection is 21.7% higher than that of the original algorithm.The root mean square error of pose is less than 0.5 meters.Therefore,it is verified that real-time 3d reconstruction on the mobile end is feasible and more efficient than the original algorithm.