Laser radar three-dimensional target rapid detection method based on pseudo-image technology

Abstract

The invention provides a laser radar three-dimensional target rapid detection method based on a pseudo-image technology. The method comprises the steps that S1, removing ground points in original point cloud data; S2, converting three-dimensional point cloud data into a two-dimensional pseudo image data form with three-dimensional feature information; S3, setting a two-dimensional convolutional neural network fusion feature layer, and capturing feature maps of a pseudo image under three different scales; S4, realizing position regression and category identification of a target by utilizing anSSD detection positioning network based on the feature maps of the pseudo-image under different scales; S5, setting a total loss function of a detection network; and S6, performing network training toobtain a standard three-dimensional target detection network for performing three-dimensional target detection based on laser radar acquisition point cloud data. According to the method, a large amount of calculation is reduced, and the position, size, shape and other information of a detected target are accurately obtained while the detection process speed is ensured.

Description

technical field [0001] The invention belongs to the field of unmanned vehicles, and relates to an environment perception method for unmanned vehicles, in particular to a fast three-dimensional target detection method based on laser radar. Background technique [0002] 3D object detection based on point cloud is an indispensable task in the development of unmanned driving technology. The control system of an unmanned vehicle relies on sensors to perceive the surrounding environment, including vehicles, pedestrians, and traffic signs. Unmanned vehicles are usually equipped with a variety of sensor devices to obtain more information. Different sensors can complement each other to provide richer information for the control system of unmanned vehicles to perceive. Lidar sensors and camera sensors are the most commonly used and also the most important sensor devices in unmanned vehicles. The point cloud data collected and generated by lidar is currently the main input to realiz...

AI Technical Summary

Problems solved by technology

However, the three-dimensional convolution operation is very computationally intensive in large scenes, and it cannot even be realized.

Images