Obstacle Detection and Tracking Method for Autonomous Vehicles in Ports

Abstract

The invention discloses a low-cost obstacle detection and tracking method for unmanned driving, which includes acquiring environment point cloud and superimposing the point cloud to obtain the environment point cloud set PC_0 of the current frame; in the vehicle body coordinate system, removing the height Points whose difference is greater than the threshold, get the point cloud set PC_1 composed of non-ground points; filter the length and width of PC_1, keep the environment point cloud within a fixed distance range, and then convert the point cloud into a single-channel or three-channel feature image; The obtained feature image is input into the built detection model to obtain the information list of obstacle targets; according to the obtained obstacle target information list, combined with the obstacle target information list of the previous frame, different methods are selected for different types of obstacle targets Or parameters to build a tracker to track each target in the current frame, and calculate and update the movement information of obstacle targets at the same time; repeat steps 1-5 until the end of automatic driving. The method is low in cost, high in detection accuracy and fast in operation.

Description

Technical field [0001] The present invention relates to the field of unmanned aerial vehicles, in particular to an obstacle detection and tracking method for unmanned vehicles in ports. Background [0002] In recent years, with the popularization of various driving assistance systems and the importance of traffic safety in all walks of life, autonomous driving technology has developed rapidly. However, fully autonomous vehicles for the public still have a long way to go, and exorbitant costs are a major factor hindering their progress. Under the existing environment perception technology, vehicles need to be equipped with several different types of sensors such as cameras, lidar and millimeter-wave radar to obtain complete and reliable road information more accurately. [0003] Although the cost of perception schemes based on monocular or multi-eye cameras is low, the optical imaging of cameras is susceptible to many factors such as light sources, scene environments, and imaging...

AI Technical Summary

Problems solved by technology

[0003] Although the cost of a perception solution based on a monocular or multi-eye camera is low, the optical imaging of the camera is easily affected by many factors such as light source, scene environment, and imaging field of view.

Moreover, the ranging accuracy of the camera attenuates sharply as the distance increases, making it impossible to obtain accurate obstacle distances

When millimeter-wave radar is used to detect obstacles, since the millimeter-wave may scatter and penetrate when it touches the target, it may sometimes cause false detection of the target and affect the detection accuracy of the entire perception system.

Although the price of lidar sensors has dropped in recent years, the price of some high-bundle lidars is still very high. For example, the unit price of Velodyne HDL-64E is as high as about 400,000 yuan, which is not suitable for the cost constraints of mass-produced models.

No reports of such techniques have been found so far

Images