Research On Navigation And Positioning Algorithm Of Mobile Robot Based On Binocular Vision

Mobile robot autonomous positioning and navigation technology is at the forefront of the current robotics field and represents the development direction of artificial intelligence technology.Simultaneous localization and mapping(SLAM)technology is a hot research topic.In recent years,with the increasing maturity of image processing technology,because of binocular sensors are closest to human perception activities,and compared with other sensors used in SLAM in the past,they have low price,simple structure,low energy consumption,good environmental adaptability and many obvious advantages such as higher accuracy and efficiency of feature depth information.It has been identified as one of the most reliable sensors for robot SLAM,so SLAM application scenarios based on binocular cameras will be more extensive.This paper implements simultaneous positioning and mapping by performing feature extraction and matching on binocular images.Based on feature extraction and matching optimization,back-end optimization,loop detection,and other improvements and verification analysis,a binocular visual SLAM scheme framework combining front-end and back-end is designed.This paper first discusses the background significance of binocular vision SLAM research and the current status of SLAM framework research,Zhang Zhengyou camera calibration method is used to perform internal and external reference calibration experiments on the binocular camera.Secondly,an adaptive feature matching algorithm based on grid motion statistics is proposed.Through the introduction of grid division to the image to be detected,and set adaptive thresholds for each grid area in order to detect feature points;and use binary descriptor rBRIEF to describe feature points and complete feature point matching based on Hamming distance;and use GMS algorithm Eliminate the first mismatched points,and use RANSAC algorithm to screen out the exact matched points,It solves the problem that the performance of GMS algorithm depends on the number of feature points and there is a mismatch set when there are fewer feature points detected.Then in the back-end optimization,a pose map optimization method is adopted to improve the back-end computing efficiency.In view of the shortcomings of common dictionary models in loop detection,an improved visual dictionary construction scheme of TF-IDF is used and loop correction is performed to increase words The degree of discrimination makes the image query more accurate.The experiment proves that the loop detection algorithm in this paper has higher detection accuracy and is superior to the traditional loop detection algorithm in loop detection performance.Finally,the actual data collected verifies feasibility in the actual positioning and construction of the binocular vision SLAM system in this subject.