An Improved SLAM Based On The Indoor Mobile Robot

Abstract

To solve the problems of trajectory drift and low precision in synchronous positioning and map building of indoor mobile robots, an improved depth SLAM algorithm is proposed. By adding scale and rotation invariant information to image features, the algorithm can obtain robust feature matching point pairs in the case of jitter and rotation. According to the set of matched point pairs, a hybrid PnP-ICP algorithm is used to estimate the pose transformation matrix. At the same time, the dictionary is built by the word bag algorithm and the image is solved by the word frequency-inverse document frequency algorithm. Similarly, loop detection is constructed to effectively eliminate the influence of accumulated errors, better solve the problem of trajectory drift, and achieve relocation in the case of lost tracking. Experiments show that the root means a square error of the translation matrix and rotation matrix is 0.043 m and 1.84 deg, and the average processing speed is 17.6 frames/s. Compared with the RGB-D SLAM algorithm in the literature, the accuracy of this algorithm is improved by about 60%, which can meet the functional requirements of realtime high precision positioning and map building for indoor mobile robots in larger scenes.