Research On SLAM Technology Of Indoor Mobile Robot Based On Binocular Vision

With the continuous development of science and technology,the direction of mobile robots has become a hot research field,among which SLAM(Simultaneous Localization And Mapping)technology has become the mainstream research method in mobile robot navigation systems.With the continuous development of technologies such as sensors and computer vision,binocular vision sensors can directly calculate the depth of objects by merging the two images obtained,which has the advantages of low cost,low energy consumption,high precision,and is widely used.In the visual SLAM system.In this paper,theoretical research on the SLAM technology of binocular vision is carried out.For indoor scene visual SLAM tracking phase,because the traditional Culling feature mis-match method uses the RANSAC(Random Sample Consensus)algorithm,the algorithm has the problems of slow calculation speed and multiple iterations when searching for the interior point set that meets the requirements of the solution model,which leads to the decrease of the real-time performance of the SLAM system,and affects the initial camera pose estimation accuracy.Therefore,this paper proposes some improvements for these shortcomings.Through the SCRAMSAC(Spatially Consistent Random Sample Consensus)algorithm,the initial feature matching set is preprocessed,which effectively improves the efficiency of the algorithm.After experimental comparison and analysis,it is proved that the improved algorithm can greatly improve the speed of operation and reduce the number of iterations.Visual SLAM loop closing phase for indoor scenes,the BoVW(Bag of Visual Word)model is usually used to classify the images.The traditional BoVW model is built based on an unsupervised clustering algorithm,usually using the K-means++algorithm.Since K-means++algorithm must select all the cluster center points in the clustering process to select the next cluster center point,to obtain K cluster center points must traverse the data set K times,resulting in a large scale The traditional BoVW model on the data set has high time complexity and low computational efficiency.Therefore,this paper proposes an improved BoVW model algorithm and uses AFK-MC~2 algorithm to construct the BoVW model.Through simulation experiments,it is proved that the improved BoVW model can effectively reduce the algorithm time complexity and improve the image classification accuracy.Finally,this paper builds a visual robot SLAM hardware and software experimental platform for mobile robots.Firstly,a four-wheeled robot is designed as a mobile robot hardware platform,and a binocular camera is used to collect scene information,and the image information is transmitted to the upper computer for processing;then,a ROS-based binocular vision ORB-SLAM2 software environment is set up;In view of the improved algorithm proposed in this paper,we tested the improvement of the camera position and pose estimation accuracy in the SLAM experiment,and the effect of the improved BoVW model in the closed-loop detection,respectively.After testing,the experimental results prove that Improve the effectiveness of the algorithm.