Research On SLAM Visual Odometer Based On Feature Point Matching

In this rapidly developing and fast-paced social environment,human daily life has become more and more intelligent.How to achieve real intelligence,mobile robots will replace human work more and let mobile robots complete tasks in the actual environment.It is necessary for mobile robots to identify their position in the environment.It can be seen that it is of great practical significance for mobile robots to complete positioning and mapping.However,the development of SLAM(simultaneous localization and mapping)technology is just suitable for application in positioning and mapping.SLAM technology introduces the theory and method into the positioning and mapping of mobile robot,so that the mobile device can complete the synchronous positioning and mapping.As the front end of SLAM system,the main function of visual odometer is responsible for positioning.The main task is to provide high-quality initial values for the back end of SLAM.At the back end,through complex calculation and analysis of image sequences,the position and posture of the robot in the environment can be determined.The advantage of visual odometer based on depth camera is that the depth information between itself and the photographed object can be obtained through the camera without complex calculation,which greatly shortens the calculation time of visual odometer and improves the efficiency.This paper makes a systematic research on the front-end module of visual odometer,and makes an in-depth research on the problems of feature extraction,feature matching,matching results and motion trajectory errors in the module.The specific research contents and improvement work arrangement are as follows:Firstly,this paper introduces the theoretical concept of visual SALM technology and the performance of each module,discusses the structural model and category of the current sensor,introduces the causes of the distortion of the image taken by the camera and the methods to eliminate the distortion,and finally completes the camera calibration experiment of the sensor in this paper combined with the sensor used in this paper and the experimental principle based on Zhang Zhengyou’s camera calibration,and summarizes the results.Furthermore,compare the three mainstream algorithms,establish the priority of ORB(oriented fast and rotated brief)algorithm,and then analyze the shortcomings of ORB algorithm.When the image scale changes greatly,the matching effect of the algorithm is poor;And feature points with large amount of data are time-consuming in matching,and there are wrong matching problems in the results after matching.An improved ORB algorithm for image feature matching is given,and the better results obtained after the final improvement are applied to the back end.In the camera pose optimization link,the key frame screening and global optimization are used to improve the accuracy of camera motion trajectory estimation and ensure the effective operation of the final camera pose judgment,Finally,the improved algorithm theory is verified by relevant experiments.Finally,the hardware equipment of the laboratory is debugged,the environmental conditions and platform required for the experiment are built,and the required function library resources are loaded.Verify the feasibility of the visual odometer system of the improved algorithm in this paper by the following two ways:(1)under the data of the public data set,compare with the data information obtained by the algorithm in this paper,verify the performance of the algorithm through data quantization,and draw a line diagram to further visually observe the performance comparison of the algorithm before and after the improvement;(2)Through the data set data,build a global three-dimensional point cloud map for practical application comparison.Experiments further verify the improvement of this algorithm in SLAM system.