An improved RGB-D SLAM algorithm based on Kinect sensor

Liang Zhang, Peiyi Shen, Jieqiong Ding, Juan Song, Jingwen Liu, Kang Yi

Abstract: This paper presents an improved frontend algorithm of RGB-D SLAM, that is, Extend-RGBD-ICP algorithm. After introducing the existing frontend algorithm and analyzing its drawbacks, we present three improvement aspects: point cloud data down-sampling, matching points' selection and elimination of cumulative error of consecutive frames splicing. From the experiments on different complexities scenarios and the evaluation on benchmark dataset, it can be seen that the Extend-RGBD-ICP algorithm can achieve both accuracy and efficiency.