Research On The Detection Of Illegal Parking Event Based On Target Recognition

With the rapid growth of economy,the amount of car ownership is also growing rapidly,a substantial increase in the number of cars directly led to the parking problem the problem of illegal parking become more and more serious.How to solve the problem of illegal parking has become a worldwide problem.With the development of digital video technology,artificial intelligence technology,it has been widely used in traffic field.The detection of illegal parking based on video image has become a hot research topic.The Video-based detection of illegal parking is a complex process,which includes obtaining video from a camcorder,vehicle identification in the video and vehicle illegally parked detecting.How to access to the target object better,how to identify the target object effectively,how to judge whether the illegal parking is the key point of the problem.Due to the complexity of the problem itself,there is no particularly mature and effective method,so it need to be further studied.After the reading and studying,the experimenting and researching of the existed results,This paper proposed a vehicle identification algorithm based on feature fusion and Back-Propagation(BP)neural network and a vehicle parking detection algorithm based on Gauss mixture model is proposed.On this basis,use the computer vision library OpenCV and C++ graphical user interface library Qt development and design of an illegal parking detection system.The research work of this paper is divided into the following aspects:Firstly,this paper proposes a vehicle identification algorithm based on feature fusion and BP neural network.The gray level co-occurrence moment and Hu invariant moments are extracted from the region of interest first;then combine these features into a new feature vector;finally,the combination of feature vectors as the input of the BP neural network is trained to get the classifier or to classify and identify the vehicle.Secondly,a new algorithm based on Gauss mixture model for vehicle violation detection is proposed.This method first uses the mixed Gauss model to obtain the first background,after a certain period of time and then use the mixed Gauss model to get second background;if the two background are different and vehicles,it is determined that it is illegal parking;if the two background are different but not vehicles,judge not illegal parking,the second background is assigned to the first background to update the background;if two background are same background,it will continue to run.Finally,it is about the designing of parking detection system and setting up of the experimental platform.This paper adopted the visual library OpenCV and C++interface library Qt as the development environment.Because they are not only cross platform and open source free,but also with high performance and high stability.Qt contains a large number of UI interface classes and components,and OpenCV provides a wealth of image processing,computer vision and machine learning algorithms,especially suitable for the design and development of video vehicle identification system.The recognition effect of vehicle recognition algorithm based on feature fusion and Back-Propagation neural network is tested.Experiments show that the method has obvious improvement in speed,accuracy and stability.