Multi-strategy adaptive lane line detection method based on stereoscopic vision

Abstract

The invention discloses a multi-strategy adaptive lane line detection method based on stereoscopic vision, and the method is based on the stereoscopic vision, and improves the robustness of a lane line detection task through the fusion of the three-dimensional depth information of a disparity map and the RGB semantic information of a two-dimensional image. In the detection stage, a road surface area is adaptively segmented by priori knowledge of three-dimensional environment information to serve as a lane line estimation area, obstacles which are not related to detection are filtered out, andthe problem of interference is solved. In the lane line extraction stage, RGB semantic information of a two-dimensional image is used, and a multi-strategy fusion mechanism and a sliding window technology in an image time domain are adopted to further eliminate interference of areas such as tree shadows, road signs and local fouling related to detection. A three-order Bezier curve model is adoptedin the lane line fitting process, and order characteristics of different lane lines can be better expressed. According to the method, the robustness of a vision-based lane line detection task is greatly improved, and the data point search speed and accuracy are improved.

Description

technical field [0001] The invention relates to a multi-strategy adaptive lane line detection method based on stereo vision in the application scene of automatic driving or assisted driving, and belongs to the related field of computer vision. Background technique [0002] With the improvement of machine learning, deep learning, computer vision and other theories, the improvement of computer hardware level and the popularization of GPU applications, the traditional automobile industry has ushered in new opportunities for development. Autonomous vehicles with environment perception, positioning and navigation, high-precision map construction, decision-making planning and advanced control technology as the core have become new research hotspots. At the same time, the advanced driver assistance system (Advanced Driver Assistance System, ADAS), which applies partial automatic driving technology, has begun to be gradually applied to road driving and other functions of assisted dr...

AI Technical Summary

Problems solved by technology

Since this method does not have an adaptive specified ROI and a fixed weight in the fusion process, when the detection process is interfered by special obstacles such as vehicles and pedestrians with the same color or the same texture as the lane line on the road, the false detection rate of this type of algorithm is often low. getting high and not even working

Compared with the traditional vision method, the lane line detection method based on machine learning or deep learning reduces the false detection rate to a certain extent and improves the robustness. very poor

The method based on lidar and visual fusion improves the disadvantages of a single sensor, but the two different types of sensors have shortcomings in the difficulty of development and deployment costs, and the calibration process of multi-source sensors is complicated and difficult

Images