Abstract: Aiming at the problem of low accuracy and poor stability of lane line detection in low illumination environment， an algorithm of lane line detection in low illumination environment based on model fusion was proposed. The improved color balance algorithm based on ALTM(adaptive local tone mapping) algorithm is adopted for data enhancement processing， which is beneficial for the extraction of lane line features. The improved Deeplabv3+model and Unet model are fused to reduce the overfitting. The segmented lane line image is obtained by instance segmentation. The experimental results show that the mean_IOU(mean intersection-over-union) values of the improved Unet model and Deeplabv3+model reach 0.625 and 0.646， respectively， which are 2% and 4.6% higher than the original model. The final fusion result increased by 0.01%. The stability and accuracy of lane line detection are promoted in low illumination environment.

Key words: low illumination environment; lane line detection; data enhancement; model fusion; instance segmentation