Vehicle Detection And Counting System Based On Deep Convolutional Network

Intelligent vehicle detection and counting are becoming more and more important in the field of traffic management,however,their detection is still a challenge due to the different sizes of vehicles and their moving vehicles,which directly affects the accuracy of vehicle counting.In addition,under the high viewing angle of the traffic surveillance camera,the size of the vehicle changes greatly,and the detection accuracy of small objects far from the road is low.This paper will use image processing and computer vision technology to detect and count vehicle objects in combination with the actual road environment.A deep convolutional neural network is introduced to solve the traditional algorithm affected by illumination changes,vehicle adhesion,background interference,etc.,and a vehicle detection and counting algorithm based on a deep convolutional network is proposed to finally realize a high-accuracy vehicle detection and counting system.The main research work of this paper is as follows:(1)Research the vehicle classification algorithm based on integrated deep learning technology.In view of the fact that most of the existing ones only focus on maximizing the percentage of predictions,the real-time performance is poor,more computing resources are consumed,and the classification category imbalance problem cannot be well handled,a vehicle classification algorithm based on ensemble learning technology is proposed.First,images are denoised by adaptive histogram equalization and Gaussian mixture model to improve the quality of collected vehicle images,and vehicles are detected from denoised images;then,operable pyramid transform and Weber local descriptors are employed Feature vectors are extracted from detected vehicles;finally,the extracted features are used as input to an integrated deep learning technique for vehicle classification.The classification accuracies of 99.28% and 99.13% are obtained on the BIT vehicle dataset and MIO-TCD dataset,respectively.(2)Research the vehicle detection algorithm based on YOLOv5 and ORB.Aiming at the problem that the size of the vehicle changes greatly under the high viewing angle of the traffic surveillance camera,and the detection accuracy of small objects far away from the road is low,a vehicle detection algorithm based on YOLOv5 and ORB is proposed.First,the road surface in the image is extracted,and the newly proposed segmentation method is used to divide it into a far-end area and a near-end area;then,the above two areas are put into the YOLOv5 network to detect the type and location of the vehicle;finally,through ORB The algorithm obtains the vehicle trajectory,which can be used to judge the driving direction of the vehicle and obtain the number of different vehicles.The experimental results verify that using the proposed segmentation method can provide higher detection accuracy,especially for the detection of small vehicle objects.Moreover,the described new strategy excels in judging the driving direction and calculating the vehicle.(3)Study the vehicle counting model.The video is preprocessed and provided to the object detection part,after detecting the location(bounding box)and category score of the object and other information,it will be sent to the counting module.The capture environment method is proposed,in order to make the algorithm more robust,it should count vehicles even if the object detection is not 100% accurate.(4)Design and implement a vehicle detection and counting system based on deep convolutional networks.Use Spring MVC as the basic framework for Java Web system development,and My SQL as the database for storing statistical results.In the development stage of the whole system,system requirement analysis,system outline design,system detailed design,system realization and function test are carried out to ensure the stability and practicability of the system..