A Method of 3D Target Detection Based on Direction Angle Information

Abstract

The invention discloses a method for carrying out three-dimensional target detection based on direction angle information, which comprises the following steps: performing feature fusion on color images and depth images collected by RGB-D cameras through Depth-RCNN and classifying them through support vector machines to obtain 2.5-dimensional detection frame and the corresponding two-dimensional segmentation instance map; select p on the obtained two-dimensional segmentation instance map 1 ,p 2 ,p 3 Three points, and optimize according to the constraints; the coordinates of the three points extracted in the three-dimensional coordinate system through the two-dimensional to three-dimensional point cloud reconstruction calculation determine the orientation angle of the initial three-dimensional frame; according to the fused features, the initial three-dimensional box for regression. The present invention improves the accuracy of direction angle estimation while ensuring the accuracy of the center and length, width and height of the three-dimensional frame through the fusion and utilization of depth information and direction angle information, laying a foundation for improving the detection accuracy of three-dimensional objects.

Description

technical field [0001] The invention relates to a method for three-dimensional target detection based on direction angle information, and belongs to the technical field of computer vision and image processing. Background technique [0002] Target detection refers to the use of image processing, pattern recognition, artificial intelligence and other technologies to detect the corresponding targets by extracting different feature information of scene targets. It is an important branch of computer vision and image processing. Most of the research on traditional target detection methods is only based on the two-dimensional optical images obtained by RGB cameras, while the real world is a three-dimensional world. When it comes to application scenarios such as robot navigation, robot obstacle avoidance, and automatic driving, two-dimensional detection lacks depth The information is limited because it cannot provide the position, size, and direction of the target in the 3D scene. ...

AI Technical Summary

Problems solved by technology

[0004] In view of the above-mentioned problems existing in the prior art, the object of the present invention is to provide a method for 3D target detection based on directional angle information to solve the problem that the directional angle information in the method for 3D target detection based on 2.5-dimensional information is not fully obtained. The use of the problem lays the foundation for improving the detection accuracy of 3D objects

Images