Study And Implementation Of Multi-License-Plate Recognition System Under Complex Background

The task of vehicle license plate recognition system (LPRS) is to analyze and process observed vehicle images, automatically recognize license plate, and do relative intelligent database management. LPRS can be widely applied to highway toll collecting, automatic parking fee charging, transportation load statistic, and illegal vehicle monitoring etc. In actual application, vehicle images, being observed by camera, to a great extent have many disturbances, such as, weather, background, corrupted license plate, and leaned images and so on. What is more, monitored vehicles are usually more than one in a picture, which makes the research of multi-license-plate recognition system under complex background much more important and significant.In this paper, we have researched the theories of image recognition technique, analyzed the requirements of the system, and achieved the designment and the realization of the critical parts, and the tests of the main system functions.In order to solve the problem of the multi-license-plate recognition under complex background, several key techniques have been studied. Some improvements and innovations are listed as follows. (1) License plate location. Firstly, based on the feature of rich in change in license plate region, confine a probable license plate existing range with a preset window, which is called rough location. Then locate the true license plate regions by mathematical morphology and other priori knowledge, which is called precise location. (2) Lean correction. Mainly use Radon transform and rotation transform, which can be separated into two steps. First, obtain the vertical rotation angle to achieve vertical correction based on the principle that vertical projection is required to get the maximum number of zero value. Secondly, obtain the horizontal rotation angle to achieve horizontal correction based on the principle that distance between the top and bottom edge of a license plate is the shortest. (3) Character recognition. Divide standardized character images into several blocks, and encode one by one. Then compare the encoding result with the one in standard character library. Which makes the Hamming distance smallest is the recognition result.The system has stronger anti-jamming ability as to vehicle license plates with lean, distortion, and contamination etc., and adaptive ability with respect to the changes of the external light intensity and the image contrast. Many Multi-license-plate images, which are collected in actual traffic scenes under complex background, have been recognized, the experimental results indicate that the proposed algorithms have higher accuracy, faster computational speed and stronger robustness.