Road Surface Marking Recognition Based On Deep Learning

Road traffic sign detection and recognition plays an important role in the autonomous driving scene.It can timely feedback current road information to autonomous vehicle and make driving judgments.In the actual scene,the image captured by autonomous vehicle is usually high-resolution long-range images,and the proportion of target in the image is small.Aiming at the problems of small area,poor position,low detection rate and poor real-time performance of road traffic signs in high-resolution images,it is difficult to transplant largevolume model based on deep learning training on embedded platform.The main contents of this paper are as follows:(1)To find and read papers related to the detection and recognition of road traffic signs,analyze the research progress of road traffic signs detection and recognition at home and abroad,and understand the advantages and disadvantages of the current road traffic sign detection and recognition algorithms,and find out the existing problems of the current road traffic sign detection and recognition algorithms.(2)The basic principles of neural network are introduced.The training methods of neural network,the convolutional operation and subsampling operation are explained in detail.Several classic convolutional network structures are studied,and the advantages of deep learning in the detection and recognition of road traffic signs are proposed.(3)To make a data set for experiments.According to the requirements,the image data is added with brightness enhancement and screening operations for data amplification.Label each image in the data set with labeling software LabelImg and divide the data set into training data set and test data set to prepare an accurate and reasonable training data set and test data set for the recognition of road traffic signs.(4)The principle and detection process of the current popular deep learning-based detection algorithms,such as Faster R-CNN,SSD,and YOLOv3,are studied.The comparison experiments of Faster R-CNN,SSD and YOLOv3 detection algorithms are carried out.The experimental results show that YOLOv3 is superior to the other two algorithms in detection accuracy(93.16)% and speed(13.2FPS).In order to further verify the reliability of the YOLOv3 algorithm,a real vehicle test of road traffic signs was carried out,and the actual vehicle test data of the algorithm was obtained,and it was found that the test speed of the algorithm did not meet the real-time requirements.(5)In view of the reasoning speed problem that occurred in the YOLOv3 real car test,the model of the YOLOv3 model was compressed.According to the designed compression scheme,the model parameters were reduced by 75% without reducing the recognition accuracy,and the memory footprint was reduced to 44.3 Mb,the speed is increased to 0.019 s / frame.The experimental results show that,compared with the traditional methods for road traffic sign detection and recognition,the deep learning-based method has higher accuracy and speed.At the same time,when the deep learning model is transplanted to the embedded platform,the practical feasibility of road traffic sign detection and recognition is improved through model compression.