Virtual-real synthesis laser radar point cloud generation method and device

Abstract

The invention discloses a virtual-real synthesis laser radar point cloud generation method and device, and the method comprises the steps: selecting a plurality of models from a predefined model library, setting the deformation postures and affine transformation of the models according to the set probability distribution, placing the generated model instances in a reasonable region of a real point cloud scene, and obtaining a real point cloud scene; and finally, resampling by using a spatial index and a ray projection algorithm to obtain virtual-real synthesized point cloud data. According to the method, the authenticity of the real laser radar point cloud is reserved, the configurability and flexibility of the virtual simulation scene are combined, the data labeling information can be automatically generated, and the problem that the real point cloud data acquisition efficiency and the manual labeling efficiency are low is solved; according to the method, a large number of synthetic laser radar point clouds can be quickly generated, the generation efficiency of point cloud data is improved, and meanwhile, more training data is provided for deep learning algorithms such as a perception algorithm.

Description

technical field [0001] The invention relates to the field of laser radar simulation, in particular to a method and device for generating laser radar point clouds based on virtual-real synthesis. Background technique [0002] Since the invention of lidar in the 1960s, it has been widely used in many fields such as reconnaissance, meteorology, transportation, etc., especially in recent years, it has been widely used in the automatic driving industry. Lidar measures the propagation distance between the laser transmitter and the target object, analyzes the reflected energy on the surface of the target object, the amplitude, frequency and phase of the reflected spectrum, and other information, thereby presenting the precise three-dimensional structure information of the target object. LiDAR typically outputs 3D point cloud data representing scanned objects of interest. [0003] The perception module is the core module of the autonomous driving algorithm, which requires massive l...

AI Technical Summary

Problems solved by technology

Existing methods usually only use one kind of point cloud data or mix two kinds of point cloud data, which will lead to problems such as insufficient data volume or data consistency.

Images