Edge SLAM System Based On Bilateral Filtering And Adaptive Thresholds

Simultaneous Localization and Mapping(SLAM)is one of the core technologies for mobile robots.Since image edge features are the plainest representation of the real scene,often representing the whole image,possessing higher overall accuracy and signal-to-noise ratio,and more robustness,research on some key techniques in edge feature-based visual SLAM systems has gradually emerged.In this paper,a complete visual SLAM system based on edge features is proposed to achieve semi-dense map construction,and the main research contents include.(1)The edge detection algorithm of Cath-Canny is proposed for the analysis of the defects of the traditional Canny operator.Firstly,the image is processed by applying bilateral filtering,which not only suppresses the noise of the image well,but also preserves the high-frequency information of the edges well;then the maximum variance between classes algorithm(OTSU)is applied to determine the high and low thresholds of the Canny operator adaptively.(2)Using edge-based sliding window local bundle adjustment(BA)optimization,the adjacent keyframe poses are locally optimized;then the scene closed-loop detection and closed-loop pose-map optimization based on random fern descriptors are introduced to globally optimize the poses,and the accuracy of the candidate closedloop frames is verified using edge features to map out the complete camera and keyframe trajectories and construct semi-dense maps.(3)Construct an edge-based visual SLAM system,build an experimental platform and run the TUM public dataset to verify the completeness and effectiveness of the system.Experiments were conducted on the TUM public dataset covering various scenes and camera motions.The experiments show that the system in this paper can effectively construct semi-dense maps of scenes and performs well in terms of trajectory accuracy for most dataset sequences,and the edge-based visual SLAM system proposed in this paper is applicable to a variety of scenes with good accuracy and robustness.