Research On Vehicle-logo Image Identification Based On Support Vector Machine

More and more new technologies are applied to the transportation field with the rapid development of moderm transportation.Intelligent transportation system can provide convenient services to the public as well as effectively improve the management efficiency of the traffic management department.The license plate is a significant identification of vehicles,and its information plays a very important role in combating crime,tracing vehicles and traffic statistics.However,some vehicles have illegal behaviors such as missing license plate,replacing,covering and using a set of CARDS,which makes it more difficult to recognize.The vehicle-logo recognition as a part of the vehicle identification system can define the vehicle's unique identity more accurately and provide reliable evidence for the investigation,punishment and escape of unauthorized vehicles as well as vehicle tracking.The main research contents are as follows:(1)Based on the color and shape features of the license plate,the paper firstly realizes the positioning of the license plate region by a series of operations such as top-hat operation,edge extraction and mathematical morphology operation.Then carrying out the rough positioning of the vehicle-logo according to the prior knowledge,we can also get the precise location image after removing the background texture.(2)The characteristics of the logo image are analyzed,and seven HU invariant moments are used to extract the characteristics of the precision positioning vehicle.In view of the influence of low illumination conditions,three new HU invariant features are studied to express the features of vehicle-logo to the maximum extent.Expanding the image source in many aspects and building the vehicle mark characteristic database based on HU invariant moments.(3)A classification algorithm based on Support Vector Machine(SVM)is proposed to identify the vehicle-logo.In order to improve the recognition accuracy of the algorithm,the Gray Wolves algorithm is used to optimize the kernel function factor C and the penalty factor g at the same time.Then we randomly selected 9 kinds of vehicle-logo for testing,and the experimental results showed that:improved HU invariant moments can obviously increase the accuracy of vehicle-logo recognition,gray wolf optimization algorithm is better than the ordinary algorithm,which solves the vehicle-logo recognition problem effectively.(4)According to the algorithms studied in this paper,a vehicle identification system is developed based on Matlab and C++.The system mainly includes license plate location,vehicle location,feature extraction and vehicle identification.The effectiveness of the recognition system is verified through the monitoring images of toll booths,which laying the foundation for the promotion of the vehicle identification system.