Lane detection network based on hybrid attention mechanism

In the research of lane detection based on deep learning, this paper proposes an ultra-fast lane line detection network based on hybrid attention mechanism. The hybrid attention mechanism network composed of channel attention mechanism and bidirectional attention mechanism is added in the lane detection network based on row classification. The network can effectively extract the structural features of long pixels like lane lines by capturing the global features of the image. And a loss function is introduced which can enhance the network convergence ability when the foreground target is small. This method improves the detection accuracy of the network while slightly affecting the speed but not affecting the real-time performance compared with the original network structure.