本研究分別以正面拍攝之單車輛與多車輛影像作為測試,拍攝環境為白天、晚上、晴天與陰天。其中單車輛拍攝了329張照片,而多車輛影像拍攝了141張287面的車牌。此外也對36部機車每部分別以7種不同角度、2種不同距離拍攝14張影像,共504張影像,以評估系統的效能。 In this paper we propose a structure and design of a multiple vehicle license plates recognition system. It can search more than one correct positions of plate in an image and cut out each single word on the plates and then utilize neural network to distinguish out the information on the plates.
In the proposed method, we use preprocess of contrast enhancement to improve the accuracy of plate location. Then we use the edge angle and morphological method to get rid of the complicated background, and utilize the symmetrical characteristics to find out the plate and define the area including characters as plate locations.
Then, we use connected-component analysis to find out the slope of plate and rotate the plate according to the angles sloped, later cut out each labeled character sequentially.
In the recognition, we use the neural network because of the ability that a large number of neurons imitate the neural network of living beings. Besides, neural network has high and getting fault-tolerant, and that is helpful to against the noisy or incomplete cut characters.
This research is tested with the inputted images conduct of single vehicle and multi vehicles, shot in the evening, fine and cloudy day. And there are 329 single vehicle images and 141 multi vehicles images, totaling 287 plates. In order to test system efficiency, we also add the plates which are shot under on viewable angles.