| With the promotion and development of the city from traditional to intelligent,the intelligent requirements for traffic streets in the city are constantly improving.In order to improve the traffic congestion in different locations and time periods,targeted intelligent traffic management platform is adopted.In the actual project,vehicle identification is an important link in the system platform.The correct identification of vehicles on the street through bayonet cameras is conducive to the system platform to build a better green wave belt timing algorithm model,and improve the travel efficiency of residents through the constructed green wave belt.However,in the real environment,light problems caused by different factors such as different weather and time will have a certain impact on vehicle recognition.Aiming at the actual application scenario,an optimized YOLOV5s-P model was put forward in this paper,which improved the recognition accuracy of the model under the premise of ensuring the vehicle recognition speed,and was applied in the intelligent traffic system management platform in Huangpi District.The main research work and results are as follows:(1)Analyzed the working process of intelligent traffic system management platform in Huangpi District,and focused on the vehicle recognition technology of traffic bayonet video.Under the influence of different time points and different environments,vehicle recognition in traffic bayonet video has problems such as target loss,insufficient feature information and diverse size.In this paper,a variety of data enhancement technologies are used to process and collect the image data set of vehicles at the bayonet,and a series of optimization measures are carried out on the YOLOV5 s algorithm model to improve the efficiency and accuracy of vehicle recognition in traffic bayonet video.(2)For the problem that the number of feature extraction network of YOLOV5 s is too large and contains a lot of redundant operations,the original CSPDarkent53 extraction network is optimized.On the premise of maintaining the original feature extraction effect,the backbone network of YOLOV5Vs-P was constructed by using lightweight Depth Sep Conv module.Secondly,in order to further lightweight the model and improve the detection speed of the model,the number of parameters and calculation amount of the model were reduced by channel pruning.After lightweight,the number of model parameters decreased by 52.8% and 79.2%,respectively,and the model size was only 1.5MB.(3)In order to improve the recognition performance of the model,different activation functions are compared and the activation functions with good performance are added to different modules of the model.At the same time,in order to alleviate the problems of omission and false detection in the detection process,the Coord Att attention mechanism that takes into account both channel and spatial information is added to make the network focus on the key part of the target,so that the object detection and recognition task can be more accurate.In order to extract more feature information from the model,CARAFE up-sampling operator and cascading connection mode are introduced in the optimization process.CARAFE up-sampling operator can improve the receptive field of the network on the premise of maintaining the calculation cost,so that the sampling points are more concentrated on the vehicle target,without being affected by the environment.The cascaded connection can further expand the receptive field,enhance the target recognition ability of the network,and make the model have better performance in the vehicle recognition task.(4)According to the requirements of the intelligent transportation system control platform,the overall improved YOLOV5s-P was used to recognize and make statistics of bayonet video vehicles,and the appropriate Webster traffic signal timing algorithm was selected and applied in the intelligent traffic management system platform of Huangpi District to improve residents’ travel efficiency. |
PDF Full Text Request