Lane Line Detection in Low Light Based on Unsupervised Image Enhancement
Existing lane detection methods have good detection performance under good illumination conditions,but their performance deteriorates sharply under low illumination conditions.Aiming at the problem that the features of lane lines are not obvious in low light scenes,this paper adds ZERO-DCE network with fusion of void convolution in the pre-processing stage of lane detection algorithm,and proposes a lane detection method based on unsupervised image enhancement network.Firstly,hollow convolution was integrated into the ZERO-DCE network to improve the capturing ability of target information.Combined with the color attributes of the lane lines,two-stage image fusion method was used to improve the lane line features under low light conditions,and then UFAST network was used to detect the lane lines.The performance analysis tests on CULane data set show that compared with baseline algorithm,the performance of the proposed algorithm was basically the same under normal lighting environment.F1 values in shadow and night environments increased by 3.5 and 2.3,respectively.
NETL
NSTL
万方数据
