Research On License Plate And Wheel Image Recognition Algorithm In Complex Environment Based On Deep Learning

Since the 21 st century,social technology has developed rapidly,more and more industries are using computers to replace human for complex or difficult tasks.In the transportation industry,car licenses need often be checked and recorded because they can get the registration information of owners and vehicles.The computer license plate detection method can extract a license plate image from a photograph containing a car,and then obtain specific digital and text information based on the image.However,the styles of the wheels of different vehicles are also different.In smaller restricted areas,the texture of the wheels can be regarded as the only feature of the vehicle to distinguish it from other motor vehicles.The traditional method mainly uses the image processing method to locate and segment the license plate wheel,and finally to identify the information.But this method is easy to be affected by the environment and cause errors,it is difficult to detect small-scale targets.Therefore,this paper uses a new type of deep learning method which use the improved Faster RCNN network to detect the license plates and wheels.This article first sorts and analyzes the traditional license plate and wheel detection algorithm,and analyzes its conclusions experimentally to draw the advantages and disadvantages of this type of algorithm.Later,I studied the Faster RCNN network for deep learning,and used this network to optimize and improve the network according to the characteristics of the target and road driving.The original Faster RCNN first used the CNN(convolutional neural network)to extract the feature map,then used the RPN network to extract the region proposals anchor regions,and the subsequent roi pooling operation redefined all the anchor regions,finally sent them to the fully connected layer for specific classification.According to the characteristics of license plates and wheels,improved the original Faster RCNN network,mainly as follows: 1.Based on the characteristics of the license plate and wheels of motor vehicles on Chinese roads,the size of the anchor frame in the proposed anchor frame network RPN is optimized to accelerate the convergence speed and improve detection accuracy of scale anchor area.2.Improve the classification and regression method of the algorithm,using the global pooling layer instead of the original full connected layer,reducing the parameters and calculations,and improving the speed of the algorithm.In this paper,the improved deep learning algorithm is tested and compared with the traditional algorithm.The results show that the algorithm is more accurate for small-scale targets than the traditional algorithm in complex environments such as long distance or insufficient lighting rate.