Research On Mobile Key Vehicle Comparision And Early Warning Algorithms

At present,the main road monitoring has been basically perfect,mostly for the fixed monitoring system.In the urban and rural areas of remote sections,video surveillance system is relatively weak,mobile robots can be deployed in these sections,automatically identify the default vehicle license plate and alarm.Mobile vehicle video surveillance system can be easily carried and deployed,while the repair is easy through robots being retrieved.Through the database comparing,timely early warning information is issued,with timely tracking or interception of key vehicles,to achieve the dynamic regulation of key vehicle drivers.The work done in this paper is as follows:(1)Using means of image processing to locate the license plate region in the mobile monitoring image,the feature of the license plate image is extracted and the classifier of the soft interval support vector machine is trained by the SMO algorithm.Positioning the character regions by means of image processing,the feature of the character image is extracted and the BP algorithm is used to train the classifier of the feedback forward artificial neural network.(2)Using the cross-validation to measure accuracy,the number of hidden nodes in the neural network and the low resolution of the character extraction feature are determined by the trial and error method.The principal component analysis is used to reduce the feature dimension,thereby reduce the sample complexity and improve the accuracy of the character recognition.(3)A system debugging and process display platform is coded,and a mobile robot is designed and made.The mobile key vehicle comparision and early warning system is tested for different scenarios.This dissertation describes the algorithm principles and process steps,and uses experiments to evaluate and optimize the algorithms for my collection of road samples.The correct classification rate of the support vector machine is 98%.The accuracy rate of the license plate detection is 90% and the accuracy rate of the license plate segmentation is 90%.The optical character recognition accuracy is 96.8% which is increased to 99.2% after the principle component analysis module is added in the feature extraction part.The product of the license plate region detection accuracy,license plate character segmentation accuracy and optical character recognition accuracy is the overall license plate recognition accuracy which is 80.4%.Experiments show that the algorithms and process in this dissertation have a good effect with high character recognition accuracy and can simultaneously identify multiple license plates in different backgrounds and in the case of light,distance and angle changes,with well performance and promotional value.