CN107085707A - A license plate location method based on traffic surveillance video - Google Patents

Abstract

The invention discloses a kind of license plate locating method based on Traffic Surveillance Video, specifically include 1, the original image in Traffic Surveillance Video is changed, original image is converted to gray level image, and original color space is converted to HSV space；2nd, the image after conversion is pre-processed, 3rd, rim detection is carried out to pretreated image, 4th, on the basis of rim detection, integrated use local thresholding method, Global thresholding and dynamic thresholding method, divide the image into black, white dichromatism region according to the relation of pixel value and threshold value and realize image binaryzation；5th, after image binaryzation, using the expansion in mathematical morphology, area filling and etching operation, realize that region merging technique and noise are further rejected；6th, floor projection and upright projection are carried out successively to the edge image detected, determines the border up and down of car plate；7th, projection carries out License Plate after confirming by SVMs.The problems such as ambient noise pattern similar with background that the present invention can overcome prior art to exist is disturbed, characters on license plate is discontinuous.

Description

A license plate location method based on traffic surveillance video

technical field

The invention belongs to the field of video processing, in particular to a license plate location method based on traffic monitoring video.

Background technique

With the development of the economy and the improvement of people's living standards, the number of cars has also increased rapidly, which has brought many challenges to urban traffic. Therefore, automatic license plate recognition and analysis technology based on intelligent image processing theory has gained more and more attention. License plate location is a crucial part of license plate recognition, and its accuracy and recall rate are directly related to the work of subsequent links and even the performance of the entire system.

Common license plate location methods include methods based on image texture features, methods based on transformation, methods based on artificial neural networks, methods based on mathematical morphology, and so on. Methods based on image texture features usually need to select an appropriate edge detection operator first, and supplemented by image preprocessing to achieve better results; methods based on transformation are often difficult to deal with blurred or deformed license plate borders, and it is also difficult to eliminate noise. Influence; the method based on artificial neural network has good fault tolerance and learning ability, but the time required for training is often longer, and the convergence and convergence speed are not guaranteed; the method based on mathematical morphology is difficult to deal with character connection, The characters themselves are not connected and so on.

Contents of the invention

Purpose of the invention: Aiming at the problems existing in the prior art, the present invention provides a method for locating license plates based on traffic monitoring video that can overcome the problems of background noise and background similar pattern interference, discontinuous license plate characters, etc. in the prior art.

Technical solution: In order to solve the above technical problems, the present invention provides a license plate location method based on traffic monitoring video, the specific steps are as follows:

The first step: the original image in the traffic monitoring video is converted, the original image is converted into a grayscale image, and the original color space is converted into HSV space;

The second step: preprocessing the converted image,

Step 3: Perform edge detection on the preprocessed image,

Step 4: On the basis of edge detection, the local threshold method, the global threshold method and the dynamic threshold method are used comprehensively, and the image is divided into black and white regions according to the relationship between the pixel value and the threshold value to realize image binarization;

Step 5: After the image is binarized, the expansion, area filling and erosion operations in mathematical morphology are used to realize area merging and further noise removal;

Step 6: Perform horizontal projection and vertical projection on the detected edge image in turn to determine the upper, lower, left, and right boundaries of the license plate;

Step 7: After the projection is confirmed, use the support vector machine to locate the license plate.

Further, the step of preprocessing the converted image in the second step is as follows: use the characteristics of the traffic monitoring video to eliminate the background of the camera, and apply pre-filtering technology to the obtained video image to further eliminate noise points.

Further, the edge detection in the third step is specifically implemented by suppressing background low-frequency features through a gradient operator, Roberts operator, Sobel operator, Prewitt operator or Canny operator.

Further, in the process of converting the original image into a grayscale image in the first step, grayscale stretching processing needs to be performed on the generated grayscale image.

Further, in the second step, the specific steps of using the characteristics of the traffic monitoring video to eliminate the background of the camera are as follows: take the previous several frames of images for one-time determination; calculate the difference between each image and the background image, and then perform Gaussian prefiltering operation.

Further, in the seventh step, the steps of license plate location by support vector machine are as follows: standardize the size of positive samples and negative samples, use support vector machine for offline training, apply to the video to be processed after training, and realize the license plate location .

Compared with the prior art, the present invention has the advantages of:

Firstly, the present invention integrates the advantages of the three types of methods based on image texture features, artificial neural networks, and mathematical morphology, and can obtain license plate location results more quickly and reliably; secondly, using The inherent nature of traffic surveillance video eliminates the background, further improving the speed and robustness of the system.

Description of drawings

Fig. 1 is the general flowchart of the present invention.

detailed description

The present invention will be described in further detail below in conjunction with the specific embodiments with reference to the accompanying drawings. It should be emphasized that the following description is only exemplary and not intended to limit the scope of the invention and its application.

The license plate location method based on the traffic monitoring video proposed by the present invention comprises the following steps:

(1) Image conversion: Convert the original image in the traffic surveillance video to a grayscale image, and convert the original color space to HSV space. Among them, in the conversion from the original image to the grayscale image, in order to avoid the problem of grayscale concentration caused by incorrect exposure, it may also be necessary to perform grayscale stretching on the generated grayscale image.

Take the common YUV video format as an example, where Y is the luminance component (representing the outline of the image), and U and V are the chrominance components (representing the color of the image). According to the existing RGB space, the value of Y can be calculated according to the following formula:

Y＝0.299*R+0.587*G+0.114*B

The weights in the formula come from the visual model of the human eye. The green component G, which is more sensitive to the human eye, has a larger weight, while the blue B, which is less sensitive to the human eye, has a smaller weight.

In the specific implementation, in order to avoid the concentration of gray levels caused by incorrect exposure, the following steps can be used to perform gray scale stretching processing on the generated gray scale image:

Step 1: Start from gray level 0, search the histogram in the direction of increasing gray level, if the absolute value of the slope is greater than 20, save this gray level as f _min .

Step 2: Starting from gray level 255, search the histogram in the direction of decreasing gray level, if the absolute value of the slope is greater than 20, store this gray level as f _max .

Step 3: The result of grayscale stretching is as follows:

Here f(x,y) and g(x,y) represent the values at coordinates (x,y) of the current image and the stretched image respectively.

(2) Image preprocessing: Use the characteristics of traffic surveillance video to eliminate the background of the camera, and apply Gaussian pre-filtering to the obtained video image to further eliminate noise points. Among them, the camera background elimination takes advantage of the characteristics of the surveillance video, and takes the previous several frames of images for one-time determination; after calculating the difference between each image and the background image, the Gaussian pre-filtering operation is performed. In the specific implementation, the first 100 images of the traffic surveillance video can be used as the training set to generate the background frame, and the Gaussian pre-filtering can be carried out with a window of 5×5.

(3) Edge detection: Use gradient operator, Roberts operator, Sobel operator, Prewitt operator or Canny operator to suppress background low-frequency features. Among them, the gradient operator is approximated by the first-order difference:

here

In order to avoid the interference of horizontal projections on horizontal boundaries, it is possible to detect only vertical boundaries:

(4) Image binarization: On the basis of edge detection, the local threshold method, the global threshold method and the dynamic threshold method are used comprehensively to divide the image into black and white regions according to the relationship between the pixel value and the threshold value, so that each A pixel is represented by only one bit.

During specific implementation, assuming that f(n) is the number of pixels with a grayscale change value of n, and n_max is the maximum grayscale change value, the following pseudo code can be used to determine the binarization threshold, where r is a preset percentage , and use a part of the edge with a larger grayscale change value as the binarized edge in the subsequent step.

(5) Mathematical morphology operation: On the basis of binarization, the dilation, region filling and erosion operations in mathematical morphology are adopted sequentially to realize region merging and further noise elimination. During specific implementation, the color of the area filling is set according to the research object. Taking the standard small car in our country as an example, it should be filled with blue to extract the license plate.

(6) Projection confirmation: perform horizontal projection and vertical projection on the detected edge image in turn to determine the upper, lower, left, and right boundaries of the license plate.

In specific implementation, the steps of horizontal projection are as follows:

Step 1: Use f(j) to denote the number of boundary points in each row, where j is the corresponding height of the row, and regard it as a horizontal projection curve.

Step 2: Perform sliding window smoothing with a width of 5 on the above horizontal projection curve. The smoothing algorithm is to take the mean value of the window of 5×1 as the value of the corresponding position on the new curve for each pixel position on the original curve, and set the new The curve is g(j).

Step 3: For each position on the new curve, if the mean value of its 5×1 neighborhood is not less than 16, then down=j, j=j+1, skip to step 4; otherwise j=j+1, continue step 3.

Step 4: For each position on the new curve, if the mean value of its 5×1 neighborhood is less than 16, then up=j, skip to step 5; otherwise j=j+1, continue to step 4.

Step 5: If up-down>20, the projection ends, otherwise j=up, skip to step 3.

The steps for vertical projection are as follows:

Step 1: Use v(i) to denote the number of boundary points in each column, where i is the corresponding width of the column, and regard it as a vertical projection curve.

Step 2: Perform sliding window smoothing with a width of 11 on the above vertical projection curve. The smoothing algorithm is to take the mean value of the window of 11×1 as the value of the corresponding position on the new curve for each pixel position on the original curve, and set the new The curve is w(j).

Step 3: For each position on the new curve, if the mean value of its 11×1 neighborhood is not less than 3 and the mean value of its right 50×1 neighborhood is not less than 3, then left=i, i=i+1 , skip to step 4; otherwise i=i+1, continue to step 3.

Step 4: For each position on the new curve, if the mean value of its 11×1 neighborhood is less than 3 and the mean value of its left 50×1 neighborhood is also less than 3, then right=i, skip to step 5; otherwise i=i+1, continue to step 4.

Step 5: If right-left<5*(up-down), the projection ends, otherwise i=left, skip to step 3.

(7) Support vector machine positioning: Standardize the size of positive samples and negative samples, use support vector machine for offline training, and apply it to the video to be processed after training to realize license plate location.

In short, the license plate location method proposed by the present invention combines the advantages of the three methods based on image texture features, artificial neural networks, and mathematical morphology, and combines the inherent characteristics of urban traffic monitoring video The background image is eliminated, and the license plate location result is obtained more efficiently and reliably, which lays a solid foundation for the subsequent license plate segmentation and license plate character recognition.

The above descriptions are only examples of implementation of the present invention, and are not intended to limit the present invention. All equivalent replacements made within the principle of the present invention shall be included in the protection scope of the present invention. The content not described in detail in the present invention belongs to the prior art known to those skilled in the art.

Claims (6)

1. a kind of license plate locating method based on Traffic Surveillance Video, it is characterised in that comprise the following steps that：

The first step：Original image in Traffic Surveillance Video is changed, original image is converted to gray level image, original color Space is converted to HSV space；

Second step：Image after conversion is pre-processed,

3rd step：Rim detection is carried out to pretreated image,

4th step：On the basis of rim detection, integrated use local thresholding method, Global thresholding and dynamic thresholding method, according to The relation of pixel value and threshold value divides the image into black, white dichromatism region and realizes image binaryzation；

5th step：After image binaryzation, operated using the expansion in mathematical morphology, area filling and etching operation, realize area Domain merges and noise is further rejected；

6th step：Floor projection and upright projection are carried out successively to the edge image detected, the side up and down of car plate is determined Boundary；

7th step：Projection carries out License Plate after confirming by SVMs.

2. a kind of license plate locating method based on Traffic Surveillance Video according to claim 1, it is characterised in that described The step of being pre-processed in two steps to the image after conversion is as follows：The camera back of the body is eliminated using the characteristic of Traffic Surveillance Video Scape, noise spot is further eliminated to gained video image application prefilter.

3. a kind of license plate locating method based on Traffic Surveillance Video according to claim 1, it is characterised in that described Rim detection described in three steps is calculated especially by gradient operator, Roberts operators, Sobel operators, Prewitt operators or Canny Son is realized to suppress background characteristics of low-frequency.

4. a kind of license plate locating method based on Traffic Surveillance Video according to claim 1, it is characterised in that described Need to carry out gray scale stretching processing to the gray level image of generation during original image is converted into gray level image in one step.

5. a kind of license plate locating method based on Traffic Surveillance Video according to claim 1, it is characterised in that described In two steps comprising the following steps that for camera background is being eliminated using the characteristic of Traffic Surveillance Video：Take above some two field pictures enter Row is disposable to be determined；Calculate after the difference for obtaining each image and background image, then carry out Gauss pre filtering operation.

6. a kind of license plate locating method based on Traffic Surveillance Video according to claim 1, it is characterised in that described The step of carrying out License Plate by SVMs in seven steps is as follows：The size of specification positive example sample and negative data, is utilized SVMs carries out off-line training, and training is applied to pending video after finishing, realizes License Plate.