Traffic Sign Detection And Recognition Of Intelligent Driving Vehicle In Natural Scene

More and more countries have begun to pay attention to the research of intelligent driving system.The relevant departments in China have published relevant documents,and put forward the pre research of relevant standards as soon as possible,and start the research on the demand of automatic driving high precision map and the demand for road facilities.Intelligent driving system has become one of the main directions in the field of automotive research.The main professional areas involved in intelligent driving system can be divided into environmental perception,intelligent control and path planning.The detection and recognition of traffic signs is one of the most important components of environmental perception.Accurate and real-time detection and recognition of traffic signs are the key and difficult points in this field.This paper focuses on the detection and recognition of traffic signs in China,and uses two different algorithms to detect and identify traffic signs.The accuracy of the detection recognition algorithm and the robustness of running in various working conditions are considered as the research points.The main research methods and research contents are as follows:(1)To find and read papers related to the detection and recognition of traffic signs,analyze the research status of traffic signs detection and recognition at home and abroad,and understand the advantages and disadvantages of the current traffic sign detection and recognition algorithms,and find out the existing problems of the current traffic sign detection and recognition algorithms.(2)Introduced the domestic traffic signs,and analyzed the characteristics of warning signs,prohibition signs and instructions.In order to increase the detection and recognition accuracy of traffic signs,image preprocessing is introduced,and the images to be detected are processed with noise and contrast enhancement.Finally,it enumerates some traffic sign detection algorithms based on color feature,shape feature and color shape multi feature fusion,which is based on template matching,support vector machine and neural network recognition algorithm.(3)Designed a traffic sign detection network.The basic structure of convolution neural network is expounded,and the convolution operation,pooling operation,the principle and function of the full connection layer are explained in detail.The structure and characteristics of cascaded convolution neural network and the structure of deep separable convolution network are studied.A deep separable cascade convolution neural network is proposed as a traffic sign detection network.(4)The identification network of traffic signs is designed.A stacking convolution network is proposed as one of the identification network of traffic signs,and its parameters are set.The structure and principle of dense convolution neural network are studied.As the depth of the convolution network is deepened,the generalization ability of the model is better.The dense network can realize the deep connection of the network.The parameters of the dense network are explained,and the dense network is used as another identification network for the traffic sign,and the network is carried out with the overlapped network.Comparison analysis.(5)In the end,the software and hardware used in the test and identification experiment of traffic signs and some traffic sign data sets at home and abroad are introduced,and some differences between domestic and foreign symbols are analyzed.The traffic sign data set of Qinghua is selected as the training set of the model,and the actual vehicle test is carried out according to the model type trained.The accuracy of network detection and recognition is compared and evaluated.The accuracy of the algorithm is evaluated by using the traffic sign data collected in the natural scene,and the two networks are used to detect and identify the traffic signs under the occlusion,backlight and rain weather conditions.The experimental results show that:(1)Compared to the traditional machine learning algorithm,the accuracy rate of the detection and recognition of the depth learning algorithm is higher,and it can detect and identify traffic signs very accurately.(2)Compared with paving network,dense network layer has more layers,deeper network and higher accuracy,but its running time cost has increased.Dense networks can accurately identify tag images in backlight,shelter and rainy days.