3D target detection method based on context perception feature aggregation

Abstract

The invention discloses a 3D target detection method based on context perception feature aggregation. The method comprises the following steps: S1, extracting a local feature point set of an input point cloud and predicting coordinates and feature offset; s2, generating an optimized search radius according to a feature cluster formed by voting operation; s3, generating semantic features according to the optimized search radius in the step S2; and S4, performing detection frame classification and regression task of the 3D target according to the semantic features in the step S3 to obtain a final output result with a 3D detection frame. Compared with the prior art, the target detection method provided by the invention has the advantages that the precision of a target detection task is improved while the lightweight class of the network is maintained, and an ideal target detection effect is achieved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a 3D target detection method based on context-aware feature aggregation. Background technique [0002] In recent years, the rise of deep learning has greatly promoted the rapid development of the field of computer vision. It has achieved outstanding results in many traditional 2D object detection tasks, but it is difficult to directly apply to a wider range of 3D scene tasks. Object detection has important research value. 3D object detection is a computer technology for locating and recognizing objects in 3D scenes. It has become one of the important research directions of computer vision in understanding 3D scenes. It is widely used in autonomous driving, intelligent robots, augmented reality and other technical fields. , has important research significance. [0003] 3D object detection based on point cloud data has received more and more attention due to its simple an...

AI Technical Summary

Problems solved by technology

The projection-based method converts data into image form, and migrates the mature 2D target detection method to the 3D field. It is limited by the 2D target detection framework, does not make full use of 3D data information and has poor scalability.

The voxel-based method can better extract local features, but it is easy to lose information when the resolution of voxelization is low, and it requires high computational cost when the resolution is high

The point-based method maintains the input data as a point representation. Compared with other methods, it has the advantages of lower complexity and higher efficiency, but its ability to capture and gather local features is weak, resulting in a decrease in recognition accuracy.

Images