Research And Implementation Of Bus Crowding Classification Algorithm Based On Deep Learning

A reasonable transportation system is the main driving force for urban development.The basic transportation mode is to transport people and logistics to large and medium-sized cities,and plays a decisive role in the flow of labor,the transportation of material resources and the layout of industrial areas in cities.Therefore,a perfect transportation system is the main artery for the rapid development of the city.At present,the modes of transportation for serving the public within the city are more perfect,and due to the popularization of green and low-carbon travel concepts,nearly 60% of the public will choose public transportation,so how to improve the city The public transportation system has become one of the key directions of government work.Bus now play an increasingly important role in public transportation,and bus companies in various cities have invested a lot of resources in the construction of intelligent bus dispatching systems.Intelligent vehicle surveillance cameras have become the basic accessories for buses.Integrated video surveillance,satellite positioning,4G communication and other functions.However,in the construction of the intelligent bus dispatching system,how to efficiently and in real-time detect the crowding degree of passengers in the bus,and dynamically dispatch vehicles for each bus line to ensure that the number of online buses in each bus line meets passenger demand and enhance passengers.The car experience without wasting bus resources has become an urgent problem.The main goal of this paper is to propose an algorithm that can detect and classify the crowdedness of passengers in the bus compartment in an efficient and real-time manner.The intelligent vehicle monitoring camera can be used to obtain the video frame data of each bus in real time.According to the algorithm proposed in this paper,the video frame data is obtained.Perform calculations to get real-time passenger congestion in the bus.The main research results of this paper are as follows:(1)It is proposed that the deep learning algorithm is applied to the classification of bus passengers' congestion degree.The traditional traffic sensor and infrared detection methods have obvious defects.The method based on image processing technology for cabin congestion detection is inefficient and accurate.The rate is insufficient,and the combination of deep learning and crowding detection scenarios proposed in this paper has achieved very good results.(2)After comprehensive analysis of the current classical deep convolutional neural network,this paper selects the GoogLeNet model as the basic algorithm for research,and because of the limitations of the GoogLeNet model,such as large scale,large computational complexity and high model complexity,it is obviously unable to meet the practical application.On the basis of this,this paper designs a lightweight GoogLeNet model to improve the image processing speed without reducing the overall accuracy of the model,and meet the real-time requirements in the actual scene.(3)Aiming at the problem that the lightweight GoogLeNet model has a poor classification effect on the 0th and 1st categories in the dataset,according to the fusion model idea,the MixNet neural network model is proposed,and the MixNet extracts the extracted bottom layer features and high-level features.It shows that the classification effect of Class 0 and Class 1 has been significantly improved.(4)According to the characteristics of the data sets used in the experiment,this paper proposes to integrate the attention mechanism based on the MixNet model.The experimental results show that the attention mechanism is integrated.The model has achieved good results on relevant indicators.