Abstract At present, the target detection network based on deep learning still has some problems in the field of lane line recognition, such as unclear Lane difference, low recognition accuracy, high false detection rate and high missed detection rate, high delay and high power consumption. In order to solve these problems, this paper proposes an asymmetric convolutional lane detection and tracking network AK-SCNNLane based on spatial instance segmentation. In the coding part, VGG16 network with asymmetric convolution kernel and spatial convolution neural network (SCNN) are applied to improve the ability of network structure to learn spatial relationships, which solves the problems of fuzzy, discontinuous and low real-time prediction of lane lines. At the same time, based on LaneNet, two branch tasks after encoding output are coupled to improve the poor foreground and background recognition and the indistinguishability between lanes. Finally, the method is compared with several semantic segmentation-based lane line algorithms in TuSimple dataset. Experimental results show that the accuracy score of this algorithm is 97.12% , and the false detection rate and missing detection rate are better than other networks. Compared with LaneNet, the false detection rate and missing detection rate are reduced by 44.87% and 12.7%, respectively. At the same time, the amount of calculation is significantly reduced, and the runtimes is reduced by 89.74% compared with LaneNet, which effectively improves the recognition and detection speed, and meets the real-time requirements (+30fps) that automatic driving scenes must meet. To sum up, the detection algorithm is not only effective, but also efficient, and basically meets the requirements of real-time lane detection and tracking.in comparison with LaneNet, the false detection rate is reduced by 44.87% and 12.7%.