Research On Detection And Recognition Algorithm Of Road Traffic Signs Based On Neural Network

Traffic signs play an important role in providing effective road information to drivers.The traffic sign detection and recognition system can provide information for the traffic department and contribute to the maintenance of road traffic signs.Traffic sign detection and recognition system is an important subtask in advanced driver assistance system and automatic driving system.How to balance the accuracy and applicability of traffic sign detection and recognition is still the difficulty and key point in this field.Therefore,this paper studies the detection and recognition of domestic traffic signs in real environments,improves algorithms,and improves the accuracy of the network and the applicability in the real environment.The main work of this paper is as follows:In this paper,we improve the Pulse Coupled Neural Network.As a single-layer neural network model,PCNN network can be directly used for image processing such as segmentation and detection,with good real-time performance.In this paper,we propose a method of parameter adjustable modified SPCNN model and apply to traffic sign detection.We improve the connection strength and connection weight amplification coefficient between each neuron in the model,and add an auxiliary parameter.Combined with the dynamic characteristics of neurons,we derive the internal activity items and dynamic threshold of the model.The accuracy of the model is verified by the test of TT-100K data set.When the simplified PCNN model is used for detection,if the edge pixels of the image are close to each other,the fitting effect of the detected image will be incomplete in some subtle parts,and the detection effect will be affected.In order to solve the above problems,the proposed detection algorithm combines the lateral inhibition network in human visual characteristics and SPCNN network,which can obtain a more refined detection effect.The anti-noise performance of the lateral inhibition network is poor.By introducing a possibility measure factor into the acyclic lateral inhibition network,the anti-noise performance of the network is effectively improved.After testing on TT-100K data set,it is verified that the proposed method has strong applicability in different real environments,such as uneven illumination and deformation of traffic signs.The traditional neural network has a large amount of calculation and the model is complex to apply,so we choose the deep separable convolution layer to replace the traditional convolution layer.We use the 1×1 point-by-point convolution to integrate the features,which greatly reduces the computation amount of the network,and is convenient to be used in the actual automatic driving or embedded platform.In this paper,based on the Mobilenet V1 network,the TT-100K data set simulation test method was adopted to evaluate the classification and recognition of traffic signs with accuracy,loss and number of parameters.The final recognition results were counted,and the recognition accuracy rate of 92.55% was obtained.