Research On Traffic Sign Recognition Algorith In Front Of Road

Target detection and pattern recognition are important research topics in the field of computer vision.The research mainly uses the machine's autonomous learning to detect whether there is a target in the image and recoginize the target.The detection and recognition of traffic signs is a fixed target detection technology based on computer vision.Its research has important theoretical and practical value,because it is an important part of road safety management,advanced driver assistance system and driverless technology,and is closely related to urban traffic safety.The traffic sign detection and recognition method based on deep learning is the key research direction in the field of machine vision because it has the advantages of fast detection speed,high recognition accuracy,and low cost.However,the traffic sign recognition algorithm under natural road still has problems such as low detection rate,high false detection rate and low recognition accuracy.So the research work of this thesis is carried out around these problems.Capturing information by vision sensors is similar to human's^access to information.The paper mainly researches on vision-based traffic sign detection.The monocular camera is applied to capture information ahead of the vehicle due to most of the signs occur in the front of the vehicle.The main work is summarized as follows:(1)Study on traffic sign detection method based on convolution network.Aiming at the problem that common detection methods are difficult to achieve real-time,This paper improves the fast convolutional neural network model by combining a number of strategies,including feature concatenation,hard negative mining,multi-scale training and proper calibration of key parameters.First,the model is pre-trained to produce negatives.Then,hard negative mining is used to add negative samples into the network.Finally,a feature concatenation strategy during multi-scale training process is employed to enhance the performance of the model.(2)Study on traffic sign detection method based on single neural network.Aiming at the problem that common detection methods are difficult to detect small targets,this paper extends the network structure of YOLOv2 and adopts a new loss function.Firstly,the normalization strategy of the bounding box width and height difference is used to replace the method of directly using the width and height of the target,in order to effectively reduce the error of small traffic sign detection.Secondly,an additional 1×1×64 convolutional layer is inserted between the first and the second layers of YOLO's structure in order to obtain a smooth change of the extracted features.Thirdly,the third and fourth layers of YOLO are replaced by inception modules to further deepen and widen the network structure as well as reducing the number of parameters.(3)Study the training and testing of the automatic detection model of traffic signs.Based on the research content(1)and(2),the simulation experiments are carried out on the datasets and the images collected by the actual scene,and the training and testing of the traffic sign automatic detection model are completed.The test results show that the model can effectively improve the detection rate and meet the real-time requirements of the system.