Three-dimensional target detection method, system and device based on self-labeling training sample

Abstract

The invention belongs to the field of computer vision, pattern recognition and machine learning, particularly relates to a three-dimensional target detection method, system and device based on a self-labeling training sample, and aims to solve the problems that real labeled data is difficult to obtain and high in cost, and a model trained by virtual data cannot adapt to a real scene. The method comprises the steps of: performing three-dimensional target detection of an input image sequence through a trained model, wherein the model training method comprises the steps that a high-quality modelis embedded into a CARLA simulator; enhancing a point cloud data sample generated by the CARLA simulator through a sampling algorithm based on laser radar guidance; and on the basis of a three-dimensional target detector VoxelNet, performing domain offset alignment by introducing domain self-adaptive modules of a voxel level and an anchor point level, and adding consistency constraints to build adomain self-adaptive three-dimensional target detector DA-VoxelNet. According to the invention, the three-dimensional target detection model trained by the virtual data can adapt to a real scene, thedetection effect is good, and the precision is high.

Description

technical field [0001] The invention belongs to the fields of computer vision, pattern recognition and machine learning, and in particular relates to a three-dimensional target detection method, system and device based on self-labeled training samples. Background technique [0002] 3D object detection can provide more detailed spatial information and semantic information, that is, the target category and the position, orientation, and occupied space volume of the object in 3D space, and has received more and more attention in recent years. Generally, 3D object detection requires a large amount of high-quality labeled data to train the model. In the field of computer vision, the cost of collecting sufficient manual labeling data is very expensive. In addition, the labeling information required for 3D target detection is more complicated and specialized. The cost of data labeling is higher than that of classification and image target detection. Therefore, It greatly hinders t...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above problems in the prior art, that is, the acquisition of real labeled data is difficult and costly, and the model trained by virtual data cannot adapt to the real scene, the present invention provides a 3D object detection based on self-labeled training samples Method, the three-dimensional target detection method includes:

Images