The Study On Application Technology For Intelligent Transportation System With Pulse Coupled Neural Network

With the fast development of road transport system and the multiple choices for the modes of people travelling, intelligent transportation systems (ITS) based on machine vision has broader applications and important commercial value in the field of traffic management. The main objective of ITS is to capture the vehicle information through the cameras on traffic roads, and through image processing and pattern recognition approach to complete the statistical behavior of the vehicle information. Currently, intelligent traffic monitoring applications focus on the regulation violation of vehicles, road congestion analysis, bayonet vehicle monitoring, toll station vehicle management and other aspects, these applications are related to image processing techniques include moving target detection, license plate location, vehicle searches, vehicle classification, etc. Judging from current study, access to vehicle information rely mostly on vehicle motion information, license plate positioning, and these methods also played a key role for vehicle information extraction in the practical application, but not all of the scenes exist vehicle motion information in the actual application processing, and the vehicle license plate information only can as a reference foe vehicle recognition, while the real-time image processing for vehicle in traffic monitoring has important application value. So to solve the problems of vehicle image real-time processing in the static scene have become a focus in this dissertation.Artificial neural network is the use of the mechanism of biological neural networks, namely through the complex and flexible connection between a large number of relatively simple nonlinear neurons to mimic intelligent information processing of the human brain. In this dissertation, learn bionics model, the third generation of neural network-Pulse Coupled Neural Network (PCNN) is introduced into the field of image processing for Intelligent Transportation. Compared with traditional artificial neural network model, PCNN has its dynamic neurons, temporal sum features, automatic wave propagation, and other characteristics of the sync pulse issuance. While the pulse coupled neural network model also has a simple, fast, easy hardware implementation, etc. Aiming to the problems of detecting static targets existing in the field of intelligent traffic monitoring, vehicle classification, vehicle searching, traffic sign recognition, this dissertation offers corresponding solutions with pulse coupled neural network models. The methods can effectively complete the processing of traffic pictures, and improve its timeliness in real-time system. Specific work of the dissertation is as follows: 1. In order to solve the vehicle target detection problem in the static traffic monitoring images, this dissertation proposes a fast target detection method which is based on UL-PCNN model using the pulse iterative average entropy information. Firstly, his method uses the UL-PCNN model and impulse iterative excitation on static images in the search box, extracting the entropy information of binary image in different iterations, then the entropy information of different period become a sequence, averaging the sequence, set the average entropy information as a judge of the picture area is the target or background basis. This method can quickly and accurately detect the vehicle target in static images, target detection for multi-vehicle, and this method is also effective for multi-scale vehicle target detection.2. To solve the problem of fast vehicle classification, this dissertation presents a edge invariant moments model based on edge detection method using UL-PCNN model, the edge invariant moments model can be used for vehicle classification. This is the first time the method of PCNN is introduced to vehicle classification system, the method utilizes UL-PCNN model morphological image processing, extracting vehicle edge information rapid and whole, and then calculate the edges invariant moments of different vehicle types, the moment is used as feature for vehicle recognition. These features go into the SVM classifier for training to generate predictive models which can complete vehicle classification. Compared with the common methods used in vehicle classification system, this algorithm can get the edge contour information of the vehicle by means of pulsed excitation rapidly, and the contour information of the vehicle is the biggest judgments for vehicle model. Experimental results show that this method can effectively carry out vehicle classification.3. In order to solve the problem of fast vehicle retrieval in the mass image database, this dissertation proposes a multi-feature fusion method which is based on multi-channel vehicle weighted PCNN model and color characterization model for vehicle retrieval. To facilitate the study of image retrieval in the actual vehicle traffic environment, the dissertation has established a vehicle image database, the database is established using a SVM classifier with HOG characteristics of vehicle to distinguish between vehicle and non-vehicle images. Conducting vehicle searches, the model of this algorithm is based on multi-channel PCNN weighted iterative pulse ignition of the vehicle, than the ignition feature is extracted, the ignition feature of the vehicle while the most intuitive color histogram information are fused to form a joint multi-feature retrieval. Since the joint multi-feature is stable and has the advantage of smaller dimensions, all vehicles using this method to retrieve fast and very efficient.4. In order to solve the problem of quickly traffic signs identify, this dissertation aims at the feature of traffic signs themselves, proposing a new method of traffic sign recognition based on local multi-channel weighted feature model with PCNN, this method can complete the traffic sign recognition quickly and efficiently, with high recognition accuracy. Traffic sign recognition plays an important role in the intelligent transportation system, especially in the field of unmanned. Traffic sign picture has its own characteristics, such as a clear classification, shape rules. For a color traffic signs, firstly, the image segmentation, based on the characteristics of different types of images, the image can be evenly divided into several blocks, also can capture block images with a certain percentage. After the image is divided into different regions, using the feature extraction method based on PCNN model extracts the ignition timing characteristics in that each area. Here the image segmentation aims at the whole image, and the ignition time signal for each segmented part describes the details information of the segmented region, so the whole image segmentation, local feature extraction and cascade can reflect the image spatial pixel information and characteristic information of PCNN. This algorithm is suitable for road traffic sign recognition.This dissertation starts with the basic principle of the pulse coupled neural networks, then analyses the field of intelligent transportation with pulse coupled neural networks excitation pulse characteristics in-depth, made a few of improved model and apply it. Through a series of experiments to prove that the proposed methods are effective in real-time static vehicle target detection, rapid vehicle classification, vehicle searches, traffic sign recognition, but also for further research in the field of image processing with pulse coupled neural network providing a reference.