Research And Application Of RGB-D Visual SLAM Fusion IPS Algorithm For Indoor Environment

Simultaneous Localization and Mapping(SLAM)technology is an essential part of robots to achieve autonomous mobility,which can solve the problems of localization and map building in unknown spaces for mobile robots.RGB-D-based visual SLAM has received wide attention due to its low cost and the ability to perceive color and spatial depth information simultaneously.The current visual SLAM system is relatively mature under an ideal environment.However,there are two major issues in practical applications.Firstly,the robustness of visual SLAM under complex lighting conditions is not strong.The second is that the cumulative drift problem cannot be solved in specific scenarios.Focusing the issues above,the main researches of this paper are shown as follows:(1)A shadow detection algorithm based on multi-scale superpixel fusion is proposed for the image shadowing problem caused by light occlusion in the environment,and the front-end visual odometer calculation method of SLAM was improved based on this algorithm.In the improved visual odometry,the shadow template of the image is first estimated from the input RGB and depth information.Then a shadow template-based anomalous feature point rejection algorithm is proposed for image feature points affected by shadows to eliminate the shadow edges and the internal abnormal feature points,and reduce the impact of shadows on image frame matching,to ensure the accuracy of visual odometry estimation in the shadow environment.Thus,the problem of poor robustness of visual SLAM system in shadow environment is solved.(2)To address the problem of accumulated drift in visual SLAM,the Indoor Positioning System(IPS)is added into the localization constraints on the basis of the traditional bitmap based back-end optimization algorithm,the MarvelMind is used as the IPS sensor.In the improved back-end optimization algorithm,an outlier suppression algorithm is proposed to remove the positioning anomaly information obtained by IPS.The proposed back-end optimization algorithm of fused IPS ensures the global consistency of the bit pose estimation and eliminates the cumulative errors,and also solves the problem of low frequency of IPS message release for real-time localization.(3)A mobile robot hardware and software platform is built,and the improved visual SLAM front-end and back-end algorithms of this paper are ported to this platform,based on which effectiveness and feasibility of the corresponding algorithms are verified.Finally the reconstruction of indoor 3D point cloud maps is realized.