Multi- license plate location method based on CNN's Intelligent Transportation Systems Research
|Course||Signal and Information Processing|
|Keywords||Multi-objective of vehicle license plate location Cellular neural networks Color quantization segmentation Color-coded templates The mask fixed with length and width|
License plate localization and recognition technology is an important link of intelligent transportation system, it can be effectively applied in many ways. At present, the mature license plate localization method is for single vehicle images, but the method to multi-objective and multi-type license plate isn’t mature enough, so research and develop multi-objective localization recognition system has great significance.In this background, I extensively read domestic and foreign literatures, compare and refer existing methods of license plate location, and take part in the "electronic police" project of Hisense. By full use of the multiple conditions such as color, edge features and veins, a new method based on CNN cellular neural networks, integrated color of multi-objective license plate location has been put forward.Firstly, introducing cellular neural networks for image edge extraction process, construct a new license plate identification edge extraction algorithm. Experimental results show that using the CNN can draw on the edge of each direction, and when other classic edge detection operators fail, CNN can also work effectively. The structure of CNN hardware circuit is simple, and it run 103 times faster than DSP chip,106 times faster than other simulation software, and it can realize image parallel processing.Then, use the HSI color type and color quantitative structure color code template. Combine this template and the mask which obey the fixed width, we get the real plate area. Compared with the single plate width judgment method, reliability and accuracy is greatly improved.At last, in character segmentation and recognition process, we use bilinear interpolation method to realize character segmentation and recognition, and finally complete the whole license plate identification process.Through lots of experiments, such as different time and sunshine conditions, different type license plate, complex background and close color license plate, show that our algorithm can effectively locate license plate, run fast and reliable, it has certain theoretical significance and practical application value to the research of multi-objective recognition system.