Fusion method, system and device for foreground point cloud and background point cloud

Abstract

The invention belongs to the field of point cloud fusion, and specifically relates to a fusion method, system and equipment of foreground point cloud and background point cloud, aiming to solve the unreasonable fusion of foreground point cloud objects and real background in three-dimensional perception and the diversity of data samples Insufficient compatibility, manual repetitive work, low efficiency of dataset generation and labeling. The invention includes: dividing the background point cloud into two-dimensional grids; extracting the coordinates of the center point of the bottom surface of the non-empty area, the average height, the maximum value and the minimum value, and establishing a two-dimensional grid feature map and a non-empty area indication map; constructing a non-empty area Empty area δ circular neighborhood, establish a flat position indication map; generate the size scaling, position offset and multiple angle values ​​of the foreground target to be placed, traverse and search to construct a directed rectangular neighborhood in each placement situation; place Foreground objects, and perform front-background point cloud fusion. The fusion of the foreground point cloud object and the point cloud background in the present invention is reasonable, greatly reduces manual repetitive work, and improves the efficiency of data set generation and labeling.

Description

technical field [0001] The invention belongs to the field of point cloud fusion, and particularly relates to a fusion method, system and device of 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 are well solved and applied in many fields. However, 3D perception tasks are full of challenges due to the increased data dimensionality. Accurate 3D information understanding is essential in safety-critical fields such as autonomous driving and drones. Depth cameras and LiDAR (Light Detection and Ranging) are common sensors for obtaining 3D information. Compared to depth cameras, LiDAR has higher accuracy and wider sensing range, making it a key component of many 3D perception systems. However, the annotation of point clouds generated by LiDAR sensors is very time-consuming, and the probability that...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above-mentioned problems in the prior art, that is, the unreasonable fusion of the foreground point cloud object and the real background in 3D perception and the problems of insufficient diversity of data samples, manual repetitive work, and low efficiency of data set generation and labeling, the present invention provides A fusion method of foreground point cloud and background point cloud is proposed, the method includes:

Images