Laser radar target detection method and device based on deep learning

Abstract

The invention provides a laser radar target detection method based on deep learning, and the method comprises the steps: obtaining a laser point cloud; the laser point cloud comprises a plurality of points, and each point comprises a three-dimensional coordinate in a first coordinate system; after the laser point cloud is projected to a plurality of sub-grids of the plane grid, effective grids in the sub-grids are determined, and grid attributes of the effective grids are determined; inputting the three-dimensional coordinate of a point corresponding to each projection point in the effective grids in the first coordinate system, the distance from each projection point to the center point of the effective grid to which the projection point belongs and the distance from each projection point to the average value point of the effective grid to which the projection point belongs into a deep learning network model to obtain expanded grid attributes; continuously inputting the expanded grid attributes into the deep learning network, and outputting a plurality of detection frames; each detection frame has confidence; and determining a target detection frame according to the confidence coefficient.

Description

technical field [0001] The present invention relates to the technical field of automatic driving, in particular to a deep learning-based laser radar target detection method and device in an intelligent driving system. Background technique [0002] With the rapid development of intelligent driving technology in recent years, environmental perception is an important part of the intelligent driving system. The obstacle detection function is the basic function in the environmental perception system. The types of obstacles mainly include pedestrians, vehicles, bicycles and other structured roads. common objects. Lidar can directly detect the real distance and size of the target, and is the main sensor in the automatic driving solution. As deep learning technology becomes more and more mature, deep learning technology is also playing an increasingly important role in the field of automatic driving, and it is gradually called the core technology in the field of automatic driving. ...

AI Technical Summary

Problems solved by technology

However, in the point coding technology, each point cloud needs to be processed, and the domain relationship of each point cloud needs to be calculated, so the calculation amount is large, and it is not suitable for the field of automatic driving that requires real-time calculation.

[0005] Raster encoding is in Yin Zhou, VoxelNet: End-to-End Learning for Point CloudBased 3D Object Detection literature, Martin Simon, Complex-YOLO: An Euler-Region-Proposal for Real-time 3D Object Detection on Point Clouds literature, Bin Yang, PIXOR: Real-time 3D Object Detection from Point Clouds literature and Yan, SECOND: Sparsely Embeded Convolutional Detection literature, the raster encoding method does not directly process point cloud data, but projects the 3D point cloud to the spatial grid In the process, the points in the spatial grid are processed again, but the processing capacity of this processing method is still very large, and it cannot achieve the ability of real-time calculation and cannot meet the requirements of automatic driving deployment.

[0006] Therefore, the existing method has a large amount of data processing and a complex structure, which cannot meet the requirements of real-time processing; and the structure is relatively complex and limited, and it is impossible to apply the network structure and research results related to the image deep learning neighborhood, and cannot achieve real-time processing in the field of automatic driving. sexual demands

Images