Lane-level precision automatic driving structured data analysis method

Abstract

The invention relates to a lane-level precision automatic driving structured data analysis method. The method comprises the steps of multi-task deep convolutional neural network model establishment, offline training, front-end platform transplantation, road driving scene analysis and output post-processing. The method has the advantages that the deep convolutional neural network is utilized to strengthen the feature extraction capability of the road identification line, the application scene is wider, and the robustness is stronger. A driving area branch is added, so that the related application can be expanded to an unstructured road. Lane boundary branches are increased, and the accuracy under the condition that congestion scenes or lane identification lines are shielded is improved. Lane structured data including attributes of lane boundary types, lane orientations and the like are increased, and enabling high-grade automatic driving perception application. Road surface identification information including road surface identification types, positions and the like is increased, and high-grade automatic driving positioning application is enabled. By using the shared convolutionalfeatures, operation resources can be shared with other visual perception modules, and the integration efficiency is high.

Description

technical field [0001] The invention relates to the field of automatic driving, in particular to a method for analyzing structured data of lane-level precision automatic driving. Background technique [0002] Intelligence is one of the important trends in the development of the automotive industry today, and vision systems are increasingly used in the field of vehicle active safety. Single and dual current view, rear view and 360-degree surround view systems have become the mainstream perception devices of existing advanced driver assistance systems. Most of the existing commercial lane assistance systems are based on the perception of visual lane markings. Most of these systems have a single application scenario and are only suitable for structured roads with clear road markings. The method can be summarized as feature extraction of lane markings, feature clustering of lane markings and fitting of lane markings. When applied, this type of system has the following disadvan...

AI Technical Summary

Problems solved by technology

When applied, this type of system has the following disadvantages: (i). The accuracy of lane marking line extraction is not high, and it is easily affected by factors such as light and road wear; (ii). For road congestion (vehicles block road marking lines), the lane The detection rate is low; (iii). The lane fitting accuracy is poor, which can be used for close-range yaw assistance, not suitable for automatic driving path planning; (iv). The output data structure is too simple, and there is no information about the lane and boundary attributes. Clearly defined; (v). Low correlation with other vision algorithms (such as target detection module, etc.), the overall algorithm efficiency is not high after fusion

Images