Foreground point cloud and background point cloud fusion method and system, and equipment

Abstract

The invention belongs to the field of point cloud fusion, particularly relates to a foreground point cloud and background point cloud fusion method and system, and equipment, and aims to solve the problems of irrationality of foreground point cloud object and real background fusion, insufficient data sample diversity, manual repeated work and low data set generation and labeling efficiency in three-dimensional perception. The method comprises the following steps: dividing a background point cloud into two-dimensional grids; extracting center point coordinates, a height average value, the maximum value and the minimum value of the bottom surface of a non-empty region, and establishing a two-dimensional grid feature map and a non-empty region indication map; constructing a circular neighborhood delta of the non-empty region, and establishing a flat position indication diagram; generating a size scaling, a position offset and a plurality of angle values of a foreground target to be placed, and traversing, searching and constructing a directed rectangular neighborhood under each placement condition; and placing a foreground object, and carrying out foreground and background point cloud fusion. According to the method, the foreground point cloud object and the point cloud background are reasonably fused, the manual repeated work is greatly reduced, and the data set generation and labeling efficiency is improved.

Description

technical field

[0001] The invention belongs to the field of point cloud fusion, and in particular relates to a fusion method, system and equipment for a foreground point cloud and a background point cloud. Background technique

[0002] With the rapid development of computer vision, 2D vision tasks such as image classification, object detection, medical imaging and object tracking have been well solved and applied in many fields. However, 3D perception tasks are fraught with challenges due to the increased dimensionality of data. In areas with high safety requirements such as autonomous driving and unmanned aircraft, accurate 3D information understanding is essential. Depth cameras and LiDAR (Light Detection and Ranging) are common sensors for obtaining 3D information. Compared with depth cameras, LiDAR is more accurate and has a wider sensing range, making it a key component of many 3D perception systems. However, the annotation of point clouds generated by LiDAR sensors...

Images