A Study Of Vehicles' License Plates Recognition Under Complex Scenes

As an important application form of target automatic recognition, vehicle license plate recognition (LPR) technology is used in some fields, such as electron charging, pass controlling and automobile stream supervising etc. Using it the traffic management's automation grade will be improved, so more and more people are attaching importance to its interrelated technology study.Regarded as recognition method, a kind of parallel Fuzzy-Neural network algorithm is applied to recognize the vehicle's character in this thesis, In order to improve the whole recognition performance of LPR, and aiming at vehicle character the LPR problem is studied under complex scenes.Three primary modules are included in the vehicle license plate recognition system of this thesis, i.e. LP's pretreatment, LP's location, character segmentation and recognition.Vehicle license plate (LP) location and character segmentation techniques are of critical importance to the vehicle license plate recognition system. An edge detection-projection feature based algorithm to locate the LP and a vertical projection-template matching algorithm to segment the characters are proposed. Firstly, the edges are detected in a gray-level vehicle image, the result of experiment shows that the speed of detecting license plate is high and the obtained contour is very legible. Secondly, the LP region is located by projection method, the tilt angle of LP is corrected by Hough transform. Finally, the character is segmented by LP segmentation algorithm, and tilt problems are solved effectively under complex scenes.A new method of later technique of LP character recognition is studied and proposed, which is constructed by the BP neural network recognition module and fuzzy controller module. For the sake of being implemented conveniently by hardware each module is independent.On the basis of theory research, some correlative arithmetic are simulated byMatlab tool and some relevant algorithms are carried out by VC++6.0 programming tool in the system. At last a LPR software platform is constructed in this thesis.The study shows that the proposed edge detection arithmetic can detect image edge rapidly, and the LP area's contour is very clear. To demonstrate the effectiveness of the proposed algorithm, extensive experiments are conducted over a large number of real-world vehicle license plates. The results report that LP area is located and LP characters are segmented accurately by the proposed location and segmentation arithmetic that have high accuracy and robustness. Compared with standard BP network, the parallel Fuzzy-Neural network has more satisfying performance. It is satisfied with the requirement of real-time LP recognition and has some theoretical and practical significance.