| With the gradual improvement of computer hardware operating ability,unprecedented breakthroughs have been made in the field of deep learning.More and more traditional industries in order to adapt to the needs of the intelligent era,choose to enter the Internet,the integration of products and intelligence.At present,in the field of unmanned driving and assisted driving,deep learning algorithm is widely used thanks to the excellent performance of deep learning technology in image processing.It can quickly identify the emergency situation on the road,help the driver to respond,and greatly reduce the loss of life and property of the public.Automatic traffic sign recognition comes into being in this context,which is the basis for realizing intelligent driving of vehicles on the road.Therefore,based on the target detection algorithm in deep learning,this paper proposes an automatic traffic sign recognition algorithm suitable for real road in China to provide technical support for automatic driving of vehicles.First of all,as the target detection algorithm YOLOV3 is not ideal for small target image recognition,an improved idea is proposed,that is,the multi-scale feature fusion theory is used to expand the receptive field of target detection,making the algorithm more suitable for small targets such as traffic signs.Secondly,the use of GHOST module to reduce the redundancy of features to ensure that the performance is not reduced,and the amount of calculation becomes less.Reduces the system memory requirement and provides the possibility for mobile terminal embedding.GIOU is used instead of IOU as the coordinate error loss function to speed up the position relationship between the prediction box and the real box and improve the prediction accuracy of the model.Finally,Mosaic data enhancement was designed,and the Anchor suitable for the data set was obtained through K-means clustering analysis.Appropriate learning rate and batch size were obtained through a large number of experiments,and the best experimental effect was selected.It solves the problem that the recognition rate of traffic signs is relatively low under the real situation.The experimental results show that the MAP value of the proposed algorithm on the TT100 K traffic data set reaches 89.7%,which greatly improves the recognition effect of traffic signs,and its FPS value is also improved.By comparing with the original YOLOV3 algorithm and other YOLOV3 improved algorithms,the algorithm in this paper has obvious advantages.It can accurately and effectively identify the traffic signs with smaller targets,which provides an algorithm idea for the further development of real-time detection of vehicle autonomous driving... |
