Data enhancement method applied to three-dimensional target detection

Abstract

The invention discloses a data enhancement method applied to three-dimensional target detection, and the method comprises the steps: carrying out the different information loss of a network through the random change of the size proportion of a generated grid-shaped mask, generating a more dense information loss grid mask when the proportion is small, and generating a more sparse information loss network mask when the grid proportion is large. The scale of the data set can be increased and key information of the target point cloud can be reserved by reasonably adjusting the grid proportion, so that the deep neural network is prevented from being unable to converge. Compared with a traditional method that the bounding box is divided into a plurality of small bounding boxes, the discrete grid mask can still keep key information of the target point cloud even if the grid proportion is large. According to the method, different data enhancement effects can be generated for the same target frame in each information loss operation process, so that the scale of the data set is remarkably improved, and the deep learning model obtained by training has better robustness.

Description

technical field [0001] The invention relates to a computer vision data preprocessing method, in particular to a data enhancement method applied to three-dimensional object detection. Background technique [0002] 3D target detection refers to determining the position of the target from the scene and identifying the category of the target, including two processes of precise positioning and classification of the target. The current 3D object detection includes camera-based 3D object detection and lidar-based 3D object detection. The camera has the advantages of low cost and rich information, but it cannot provide accurate depth information. It is difficult for the current vision-based 3D target detection method to achieve similar performance to the lidar-based method; and the lidar-based target detection method is: The lidar sensor quickly acquires large-scale, unstructured, and disordered three-dimensional point cloud data by performing three-dimensional scanning sampling on...

AI Technical Summary

Problems solved by technology

On the one hand, excessive deletion of point cloud information will make it completely impossible to distinguish the boundary between the background point cloud and the target point cloud, making the deep neural network regard the target point cloud with insufficient information as image noise; on the other hand, excessive The retention of point cloud information will leave some objects completely untouched, resulting in less robustness of deep neural networks

Images