| Summary: | 碩士 === 北臺灣科學技術學院 === 機電整合研究所 === 99 === Abstract
License plate recognition system has been widely used in a variety of application, such as stolen cars finding, parking lot management, traffic information and so on. A new recognition system is provided in this thesis to achieve the function of real-time and mulltiple license plate recognition. The method scans and analyzes the characteristics of horizontal image lines in a regular interval, then use the mathmetical morphological algorithm to detect the region of license plate.
License plate recognition is divided into four stages: pre-processing, license plate locating, character separating and characters recognition. At first, the color digital image is converted into gray level image, then we removed the background light (may be caused by reflective). In the second stage, we scan the horizontal lines in regular intervals to obtain the edge information. From the information we can determine the possible sub-blocks of license plate and then merge the connected sub-blocks into a full license plate block.
In the stage of license plate characters separating, firstly, a binarize operation is applied to the license plate block, then a simple morphological erosion and dilation is utilized to filter out non-characters(noises). The characters can be extracted by simple tracking algorithm. The detected license plate block should be extended if the number of characters is less than five or six.
Finally, a simple geometric matching statistics algorithm is provided for fast and effective identification of license plate characters. From experimental results, we have found that the proposed method can be real-time executed on a normal PC, and the correct classification rate is more than 90%.
|
|---|
