Method for realizing scene structure prediction, target detection and lane level positioning

Abstract

The invention discloses a method for realizing scene structure prediction, target detection and lane level positioning, and relates to the fields of automatic driving, deep learning, computer vision and the like. The method comprises the following steps: firstly, constructing a neural network for lane-level positioning, scene structure prediction and target detection, and constructing a loss function mathematical model through loss between a scene structure prediction value and a target detection prediction value and a true value thereof; making a data set through an image and a map, and training the network; deploying the network on an vehicle to output a detection result; and finally, carrying out retrieval matching on the output scene structure and a map through a matching method, correcting the positioning error of the vehicle, and realizing lane-level positioning. According to the network, a data set can be made through images and maps, closed-loop training is carried out on the network, and scene structure prediction, a target detection function and lane level positioning can be completed only through image information and map information. The road structure contained in the scene structure prediction result can be used in automatic driving.

Description

technical field [0001] The invention belongs to the fields of automatic driving, deep learning, computer vision and the like, and is a neural network for lane-level positioning, scene structure prediction and target detection. Background technique [0002] With the development of deep learning, autonomous driving technology has become more and more mature, and the car's ability to perceive the environment has gradually improved. Most of today's autonomous driving platforms still use some powerful sensors (such as lidar, precision GPS, etc.) for environmental perception solutions, but most of these sensors are expensive and bulky. If only visual sensors are used to complete environmental perception tasks, it will be greatly cut costs. At present, most of the GPS used for positioning are prone to deviation or inaccurate positioning due to their low accuracy, and GPS is still unable to achieve accurate positioning in tunnels or remote areas with poor or no signal. The present...

AI Technical Summary

Problems solved by technology

[0003] To sum up, the problems existing in the existing technology are: most of the current GPS positioning accuracy is not high and cannot complete the precise positioning of tunnels or poor signal scenarios

From the perspective of network adaptability, the patent CN111047630A has poor ability to adapt to unfamiliar scenes, because if the environment changes slightly, the target detection network needs to label a large number of new data sets for training

Images