Research On Lane Detection Method Based On Deep Learning

Lane detection is the important task of automatic driving vehicle environment perception,which directly affects the control decision of vehicle execution module,and has important research significance for automatic driving.At present,most of existing lane detection method detect the lane and its number through clustering fitting algorithm,but this method has poor real-time performance and is difficult to deal with complex road conditions;In addition,the lane detection model has a large number of parameters,so it is necessary to prune the trained model,while the traditional pruning methods only focus on the compression rate of the model,which is difficult to ensure the accuracy of the algorithm in the process of pruning.To solve the problems,this paper proposes a lane detection and model pruning method based on deep learning,which realizes the efficient and accurate detection of lane lines under complex road conditions.The main contents of this paper include the following three aspects:（1）Image lane detection method based on attention mechanism.The algorithm divides lane detection into three subtasks,which includes lane number prediction,semantic segmentation and instance segmentation of lane detection model.In the model structure design,the encoding and decoding architecture is adopted,and the attention mechanism module and void convolution are introduced into each layer of the network to save the spatial position information of the image.The lane fitting is realized by clustering algorithm.The experimental results show that the lane detection method based on attention mechanism can detect the lane line in the image efficiently and accurately.（2）Pruning method of lane model based on objective optimization.Based on the lane detection model designed above,a pruning method based on objective optimization is proposed.Taking the variation of model parameters and the variation of model accuracy as the objective,the optimization objective function is established,and the optimization model under different conditions is obtained by adjusting the two weight coefficients.The model can adjust the optimization objectives according to the requirements of different application scenarios.The experimental results show that the model pruning method based on objective optimization can effectively compress the model,improve the response speed of the model and ensure the accuracy of the model.（3）Lane detection test.The road scene data set is made,including a variety of typical road scenes.The accuracy of lane detection algorithm is tested in these scenarios.The experimental results show that the lane detection model has good robustness and can effectively realize the lane detection task in a variety of complex scenes.