| Nowadays, more and more family have their private car, In 2009, the car in china almost reach 76,190,000, grow 17.8% than last year, and private car is 26,050,000, grow 33.8% than last year. Two of a hundred people have their own car in mean level. And today there are city of people more than 10 millions including SHANGHAI, GUANGZHOU, BEIJING, TIANJIN, CHENGDU, SHENZHEN, DONGGUAN, FESHAN, HANGZHOU, NANJING, SHENYANG, WUHANG, CHONGQING, XIAN, it is not a good phenomenon. In U.S there are 260,000,000 car on the road. Following the step of society, the data will grow up, and in mean while the traffic will out of burden. We will spend more time in the road, and do harm to economy. But the source of road and hands are limited, it is difficult to make it, so it is will be proved a good way to use computer and technique to solve the problem.In 1995 U.S DEPARTMENT OF TRANSPORTATION make a plan called Intelligent Transport System. Including seven field and twenty-nine department and make the plan until 2005.License plate recognition system (LPR) is very important part of ITS.Most country government pay much attention to area. There are several solutions to LPR. The first bases on changing of texture. The second base on the deference of pixels. The third base on color of car plate. The forth base on horizontal and vertical scanning. And the fifth base on wavelet technology. Actually we usually use more than one of them to make multiple plans.These articles use the method of bilateral and wavelet solution based on texture changing.The bilateral filter was creating on the base of Gauss filter, the gauss ion filter is one of frequency domain filter. The gauss ion filter is important because gauss ion function is real both on time domain and frequency domain. The bilateral filter build another gauss ion function beyond the gauss ion filter, it use the difference of the center pixel and the pixels around to be the parameter of gauss ion function., so when use this model convolution, at the moment meeting edge, the added gauss ion function will restrain the old one, and the edge will remain.Two dimension wavelet transform can divide image to four direction channels: the low frequency, horizontal, vertical, diagonal. And the statistics of the four part reflect the peculiarity of texture of the plate domain, and wavelet make sure the situation.The step of plate detection is:At the first we make a model of car plate, cut several plate model from car photo, make them wavelet transform to four direction channels, make a 8_d array by statistics with mean and standard value. And make it a kernel of a 8-D Euclid Space. The second transforms a car photo with the same way, use a sliding window go through all four channels, calculate the data inside the window and make a 8-D array with the same way of car plate model. and the third calculate the distance of the sliding window array and the model kernel, when the distance is short enough we could consider the point site of the sliding windows is inside the area of plate area.Now we have the rough site of the car plate, and on this step we use canny operator to find the edge of the car plate, and at final step we use method of least square to fit the edge to a line.Use the method we can get a good car plate divide even the license is lean. How to enhance the discrimination on complex background. and how to choose the parameter of operator, and how to make the threshold are both the research objective... |
PDF Full Text Request