Abstract:
The reliability and stability of lane line detection is very important for intelligent driving system. Because lane lines are easily disturbed by complex situations such as light, occlusion and aging, the traditional semantic segmentation network can not accurately learn the detailed features of lane lines. In order to solve this problem, this paper first introduces the CA coordinate attention mechanism in the coding network to further enhance the ability of the network to extract lane lines, then introduces the pyramid hole convolution module in parallel with the RESA module in the feature aggregation network to enhance the receptive field of the model, so as to enrich and extract the global spatial feature information, and finally samples the fused feature map to the size of the original image through the decoding network, The location and probability distribution of each lane are predicted. Experiments show that the algorithm proposed in this paper has high accuracy on the culane data set, and the comprehensive accuracy of multiple roads reaches 76.2%. The real vehicle test shows that the detection frame rate of the algorithm is 30 fps, which can carry out real-time detection in complex traffic scenes, and has high generalization and robustness.