Summary: | 碩士 === 建國科技大學 === 自動化工程系暨機電光系統研究所 === 96 === At presently, the most recognition systems can recognize only cars. Actually, vehicles in an image are multi-target. Thus, this objective of this thesis is to develop a recognition system for multi-target vehicle license plates which can recognize multi-target vehicles to improve the disadvantages of the previous technologies.
The system comprises three modules such as license plate location, character segmentation and character recognition. The license plate location applies horizontal differential which is one of the techniques in boundary detection to eliminate irrelevant noise taken by the camera. And the license plate location is capable of precisely locating license plates based on rough detection and detailed detection. In character segmentation stage, inverse is used to exactly tell the colors of different license plates. Then the character segmentation precisely extracts each character by adjusting the angle and segmenting the character based on the contour of each character. The character recognition uses back-propagation neural network to memorize the value of each character feature by inputting the character samples, and therefore it correctly identifies the characters through the computation of competitive transfer function.
In the experiment for single target, two hundred sixty-four images are captured. Of those, one hundred ninety-eight images are successfully recognized, fifty failed. The rate of identification is 79.83%. In the experiment for multi-target license plates, eighty-four images which include eighty four vehicles are captured. Of those, sixty-one vehicles are successfully recognized, twenty-three failed. The rate of identification is 72.62%.
|