Parallel Shortened Spatial Attention Module for Effective and Precision Lane Detection

With the development of computer vision, more and more systems for autonomous driving are adopting deep learning technology. Among them, lane detection aims to avoid accidents caused by the cars that deviate from their driving lanes. The lane detection task is challenging due to complex scenes and few features of distorted lane lines. Therefore, collecting the useful spatial information of the feature map related lane line becomes an important task for lane line detection. There are some spatial enhancements in feature maps, such as the spatial convolutional neural network (SCNN) [1] and the parallel spatial attention network (PSAN) [2]. To avoid computation in computing spatial attentions from top-to-bottom, left-to-right, bottom-to-top and right-to-left processes., in this paper, we design a more efficient detection system based on the PSAN concept, we shortened the spatial attention ranges, where the module only collects spatial local information and passes to the adjacent feature to reduce the computation time and enhance the lane detection performances. Simulation results show that the proposed parallel shortened spatial attention module can achieve effective and precision lane detection.