Research On Real-time Traffic Sign Detection Algorithm Based On Deep Learning

In recent years,with the progress and development of modern artificial intelligence and automated intelligent transportation systems,the driver’s driverless and assisted driving systems have provided great convenience for the driver to drive the car.Such a system can help the driver obtain timely access The latest road condition information prompts the car driver to make correct judgments and operations,so as to avoid road traffic accidents.Traffic signs are an important part of road condition information in unmanned driving and assisted driving.In recent years,with the rapid development of deep learning,there have been new solutions to the identification of traffic signs.This article has conducted in-depth research on convolutional neural networks and target detection algorithms.The purpose is to quickly detect and recognize traffic signs in natural scenes.Since the detection of traffic signs requires the algorithm to be both accurate and fast,this paper proposes a one-stage target detection algorithm Slim CSYOLO based on pruning and attention mechanism.On the basis of YOLOv3,this paper integrates multi-scale high-level semantic information through the spatial pyramid pooling module.Use channel pruning and layer pruning to compress the YOLOv3-SPP model to improve the real-time nature of its reasoning.Finally,on the basis of the compressed model,a compound attention mechanism is added to enhance the learning ability of the model.Among them,the main work and innovations are as follows.The main research work and results are as follows:(1)This paper proposes a one-stage target detection algorithm Slim CSYOLO based on pruning and attention mechanism.Use pruning-based model compression method.By judging the weight change of the BN layer in the training process,it is sparsed.On the basis of channel pruning,layer pruning is used to further compress the model size.After effective compression,the model parameters are only 1/3 of the original model parameters,which improves the inference speed of the algorithm on the basis of ensuring the accuracy of the model inference.(2)The CBAM layer,which uses the attention mechanism including spatial and channel attention,redistributes the weights of the feature distribution of the compressed model,effectively uses non-local feature information,and improves the compression model without losing real-time detection performance.The detection accuracy.In the end,this paper obtains the Slim CSYOLO model,which can meet the needs of traffic target detection in natural scenes.(3)The experimental results show that the proposed algorithm m AP and its running speed are both higher than the traditional traffic sign detection algorithm.The recognition algorithm proposed in this paper meets the real-time requirement of traffic sign detection and recognition.