An ADMM algorithm-based update point cloud registration pipeline for autonomous vehicles

• Main Authors: DING WENDONG, YU LI, SONG SHIYU, YUAN PENGFEI, HE RUNXIN
• Format: Patent
• Language: Chinese; English
• Published: 23.03.2021
• Subjects: ANALOGOUS ARRANGEMENTS USING OTHER WAVES; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; GYROSCOPIC INSTRUMENTS; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION ORRERADIATION OF RADIO WAVES; MEASURING; MEASURING DISTANCES, LEVELS OR BEARINGS; NAVIGATION; PHOTOGRAMMETRY OR VIDEOGRAMMETRY; PHYSICS; RADIO DIRECTION-FINDING; RADIO NAVIGATION; SURVEYING; TESTING

The invention provides an ADMM algorithm-based updated point cloud registration pipeline for an autonomous vehicle. In one embodiment, a system and method for point cloud registration of a LIDAR poseof an autonomous driving vehicle (ADV) is disclosed. According to the method, the pose of the point cloud with relatively high confidence in a data capture stage is selected as a fixed anchor pose. The fixed anchor points are used to estimate and optimize the pose of the non-anchor points during point cloud registration. The method may divide the point cloud into blocks to perform an ICP algorithmin parallel for each block by minimizing a cost function of a bundle adjustment equation updated with regularization terms. The regularization term may measure a difference between a current estimateand a previous or initial estimate of the pose. The method may also minimize a bundle adjustment equation updated with regularization terms when solving the pose map problem to merge optimized posesfrom the blocks to establish