Online static point cloud map construction based on 3D point clouds and 2D images

With the development of science and technology, robots have been applied to many fields to free people's hands. Environment perception and map construction are one of the key technologies for robots to achieve autonomy. In this paper, a system based on 3D point cloud and 2D image fusion is proposed to solve the problem of dynamic object segmentation and static map construction during robot motion. Different from the existing methods, the current relatively mature target detection method is used to design the extrinsic parameters between the two coordinate systems of the images and the 3D point cloud, and the probabilistic method is used to reduce the error. The above calibration results are applied to map the image detection results to the 3D point cloud to improve the segmentation accuracy of the targets. At the same time, target tracking and filtering methods are used to classify 3D points as static and dynamic. The segmented dynamic points can be applied to obstacle avoidance, while the static points are applied to the construction of a 3D point cloud map. Finally, the open-source datasets KITTI and DAIR-V2X are used to verify the proposed method, and the results show that the method is feasible and superior.