An adaptive pose tracking method based on sparse point clouds matching

Abstract

An adaptive pose tracking method based on sparse feature point clouds matching is proposed to solve the problem of ICP estimation on quadrotor. In the ordinary environment, ICP is calculated by matching the sparse feature point clouds. After the sudden loss of the depth value, the p3p method is used to obtain the world coordinate of the sparse feature point clouds and then complete the ICP calculation, which solves the problem that the traditional RGB-D SLAM algorithm is difficult to estimate camera motion when the depth loss occurs. It is compared with RGBD-SLAM v2 algorithm on TUM public datasets and the performance is verified.