Research On Algorithm For Vehicle Detection And Recognition System Based On Convolutional Neural Networks

With the rapid development of China’s economy and the continuous improvement of people’s living standards,cars have become the preferred mode of transportation for daily travel.While bringing convenience to people’s lives,they also pose challenges to the supervision and management of road traffic safety.The traditional manual management method for supervising road vehicles not only increases labor costs,but also has low management efficiency.Real time supervision of vehicles on the road using intelligent management systems has become an important way in modern traffic management.At present,vehicle detection and recognition technology still faces many problems such as low detection efficiency.At the same time,vehicles may be obstructed by obstacles such as trees and billboards while driving on traffic roads,resulting in the loss of tracking targets and unsatisfactory tracking results.Therefore,in response to the above issues,this article mainly focuses on key technologies such as vehicle detection and positioning,license plate attribute recognition,and vehicle tracking as research directions.Combining with practical application scenarios,the original detection algorithm has been improved,and the detection efficiency has been greatly improved,achieving good results,At the same time,an intelligent traffic management system was designed and implemented through software to intelligently supervise vehicles driving on the road and automatically obtain vehicle attribute information.The specific research content and achievements of this article are as follows:(1)To solve the problem of low efficiency of vehicle detection and recognition in natural scenes,this paper proposes a vehicle detection algorithm based on deep learning,which uses an improved convolutional neural network model to extract features of vehicles,conducts multi-scale feature fusion according to the size of the Receptive field of the convolutional layer,and removes redundant parameters by appropriate pruning,By designing a loss function,the problem that the small target vehicle in the distance of the road cannot be detected is solved.At the same time,a convolutional neural network model is used for license plate recognition to correct the image of license plate tilt caused by camera angle.In the corrected image,an end-to-end neural network is used for recognition.Through experiments,it is confirmed that the algorithm used in this article meets practical requirements in real-time and accuracy.(2)In response to the problem of target loss in vehicle occlusion tracking,this paper uses a Kalman filter to predict the next position of the target vehicle as the predicted value,and uses the detection network’s detection results of the next vehicle as the observed value.Based on the similarity of the two features,the Hungarian matching algorithm is used to match the vehicle for real-time and effective tracking.(3)Based on the algorithm proposed above,this article independently designed and developed a vehicle detection and recognition system in natural traffic scenarios.The system software includes multiple functional modules,and the design architecture of the software is mainly divided into front-end user interaction interface,back-end algorithm services,and data information storage.The user interface has good human-machine interaction performance,and the system can automatically detect and identify input video data,achieving intelligent road traffic management,At the same time,it also provides strong data support for the road traffic supervision department to analyze and solve traffic problems.