CN105005766B - A kind of body color recognition methods - Google Patents

Abstract

The invention discloses a method for identifying the color of a vehicle body, which comprises the following steps: a detection unit performs motion detection on input video data, and removes window glass and vehicle shadow interference areas by using the method of the difference between the maximum value and the minimum value of RGB space; The ratio of the number of pixels to distinguish between colored vehicles and black, white and gray vehicles; use the H space histogram to identify 7 colors of red, orange, yellow, green, blue, blue, and purple, and use the V space histogram and the color based on the fan-shaped area The voting method identifies three colors: black, white and gray. The present invention adopts the vehicle body color recognition strategy based on RGB color space channel difference to identify the vehicle body color, can effectively remove the interference area of the vehicle body color recognition, and improves the correct rate of vehicle body color recognition.

Description

A car body color recognition method

technical field

The invention relates to image processing technology, in particular to a vehicle body color recognition method.

Background technique

In the current intelligent transportation system, with the increase of the number of vehicles, the traffic environment becomes more and more complex, and the identification of vehicles only by the license plate can no longer meet people's needs. Vehicle color information is more likely to arouse people's interest, so as to make up for the lack of license plate recognition caused by the phenomenon of multiple license plates for one vehicle, and it is of great significance to the identification and search of vehicles, and to improve and enhance the functions of intelligent transportation systems.

At present, there are mainly two ways to identify the vehicle color: the first is to identify the overall color of the vehicle, and extract the color features of the vehicle's region of interest for identification. Firstly, the foreground image of the vehicle is obtained through target segmentation, and then the connected area is analyzed to delete the interference areas such as wheels and mirrors to obtain the area with obvious color of the vehicle. In the classification stage, a two-layer classifier based on support vector machines is used to classify colors into black, white, gray, red, yellow, green, blue and other types, but this method is susceptible to interference caused by vehicle shadow color and window glass color , the classification accuracy of the body color is low. In addition, the vehicle's overall color recognition can extract the microscopic characteristic value of each pixel of the vehicle body color by using three channels of the HSI color space (that is, a color model composed of Hue, Saturation, and Intensity), Define the threshold range and relationship of colors, and finally classify the colors with the help of K-nearest neighbor method, artificial neural network and support vector machine. The second body color recognition algorithm is to locate the position of the license plate first, and then extract the corresponding area above the license plate as the body color recognition area and perform body color recognition. However, when dealing with unrecognizable license plate videos, this type of method cannot handle the body color, which is difficult to meet user needs. In short, most of the current recognition methods are still unable to overcome the influence of vehicle shadow color and window glass color on the vehicle color recognition results.

Therefore, how to develop and design a body color recognition method that can overcome the vehicle shadow color and the color of the window glass has become one of the technical problems that need to be solved urgently.

Contents of the invention

The purpose of the present invention is to provide a vehicle body color recognition method to solve the problem of large interference in color recognition in the prior art so as to more accurately recognize the color of the vehicle body.

The vehicle body color identification method that the present invention proposes, comprises the following steps:

Acquire the video image of the vehicle and set the trip wire, perform moving target detection on the input video image data, obtain a binarized foreground moving target image, intercept the circumscribed rectangular area of the foreground moving target, and extract the foreground in the rectangular image The color pixel value corresponding to the moving target area in the original video frame, and then calculate the difference between the maximum value and the minimum value in the RGB color channel of all pixels in the circumscribed rectangular image, and remove the window glass and vehicle shadow interference area by threshold segmentation method ; Calculate the ratio of the number of pixels to distinguish between colored vehicles and black-and-white and gray vehicles. The ratio of the number of pixels is the pixel in the body area of the foreground moving target after the RGB color channel maximum value and minimum value difference processing and binarization processing The ratio of the number of pixels to the number of pixels contained in the circumscribed rectangular image of the foreground moving target;

Use the H space histogram to identify seven colors of red, orange, yellow, green, cyan, blue, and purple; use the V space histogram and the color voting method based on the fan-shaped area to identify the three colors of black, white, and gray identify;

Output the recognized body color.

The difference between the maximum value and the minimum value of the RGB color channel is the maximum value Max(R, G, B) of the R channel, G channel and B channel in the RGB color space for each pixel in the body color recognition image and the minimum value Min(R,G,B) (ie Max(R,G,B)-Min(R,G,B)).

The number of pixels contained in the circumscribed rectangle of the foreground object is based on the VIBE (Visual Background Extraction) algorithm to detect and obtain the moving object in the foreground area. If there is the first intersection with the tripwire pixel point set on the image, the number of pixels contained in the circumscribed rectangular image of the foreground moving target pixel point set is calculated.

The H-space histogram is the H-space histogram of the vehicle body area corresponding to the color area in the original video frame obtained after the vehicle image is subtracted from the maximum value and the minimum value in the RGB color channels and subjected to binarization processing.

The color voting method based on the fan-shaped area is to convert the RGB color space of all pixels in the circumscribed rectangular image of the foreground moving target passing the tripwire to the HSV color space, find the centroid of the foreground moving target, and then take the centroid as the center of the circle, Draw a circle with the shortest distance from the center of mass to the edge of the vehicle’s circumscribed rectangular frame as the radius, divide the disk area into 5 sectors at 72 degrees, and calculate the pixels in each sector area (excluding background area pixels) in V space Histogram, detect the color corresponding to the position of the highest peak in the histogram in the V space, this color is the color of the fan-shaped area, and finally the color with the largest number of fan-shaped areas is the body color of the vehicle.

In a specific embodiment of the present invention, described vehicle body color recognition method comprises the following steps:

Step S110. Obtain the vehicle video on the road to be detected;

Step S120. Use the VIBE (Visual Background Extraction) algorithm to detect moving objects on the video data to obtain a binarized foreground moving object image, extract the foreground moving objects passing through the tripwire, traverse the edges of the foreground moving objects, and record vertically and horizontally respectively. According to the point coordinates corresponding to the maximum and minimum values in the direction, determine the circumscribed rectangle of the foreground moving target according to the obtained point coordinates, and intercept the circumscribed rectangular image P1, and then extract the foreground moving target area in the rectangular image P1 in the original video frame. Corresponding color pixel value and get image P2;

Step S130. Calculate the maximum value Max(R, G, B) and minimum value Min(R, G, B) of each pixel in the RGB color channel in the circumscribed rectangular image P2, and then calculate the difference between the maximum value and the minimum value to obtain RGB color difference image P3;

Step S140. Given a threshold M1, use the threshold to perform threshold segmentation on the RGB color difference image P3 to obtain a binarized body area image P4;

Step S150. Count the number of pixels N _P4 included in the body area in the image P4, count the number N _P2 of all pixels in the circumscribed rectangular image P2 of the foreground moving object, and calculate the ratio of the number of pixels R=N _P4 /N _P2 ;

Step S160. By performing threshold judgment on the ratio of the number of pixels, vehicles are divided into colored vehicles and black, white and gray vehicles;

Step S170. When the vehicle is a colored vehicle, use the H-space histogram to identify seven colors including red, orange, yellow, green, blue, blue and purple;

Step S180. When the vehicle is a non-color vehicle, use the V space histogram and the color voting method based on the fan-shaped area to identify the three colors of black, white and gray;

Step S190. Store or output the recognition result.

The present invention extracts the foreground moving target from the video, and can save the circumscribed rectangular image of the moving target passing through the trip wire, by calculating the maximum and minimum difference and threshold value segmentation of the R channel, the G channel and the B channel in the RGB color channel, Remove the windows and shadows, and then classify the vehicles into two types: colored vehicles and black-and-white gray vehicles for judgment. If the judgment result is a color car, it will be identified according to the hue space histogram; if it is judged as a black and white gray car, it will be identified according to the V space histogram and the color voting method based on the fan-shaped area. The invention solves the problem of relatively large interference in the candidate area selected for color judgment in the prior art through the RGB channel difference method, and improves the correct rate of body color recognition.

Description of drawings

Fig. 1 is a schematic flow sheet of a preferred embodiment of the present invention;

Fig. 2 is a video image of a vehicle to be identified in the present invention (the white dotted line is a trip wire);

Fig. 3 is a color space histogram of the present invention.

Detailed ways

The invention discloses a vehicle body color recognition method, which comprises the following steps: firstly read the initial frame of the video and set the trip line, then detect the moving target on the input video data, calculate the circumscribed rectangular area of the foreground moving target and intercept the Get the circumscribed rectangle image, then calculate the difference between the maximum value and the minimum value of all pixels in the RGB color channel in the circumscribed rectangle image, and divide the vehicle color into color vehicles or black, white and gray vehicles, and finally calculate the histogram of each color Graph distribution to identify body color. The present invention adopts the vehicle body color identification strategy based on the difference between the maximum value and the minimum value of the R channel, the G channel and the B channel in the RGB color space to identify the vehicle body color, which can effectively remove the interference area of the vehicle body color identification and improve the accuracy of the vehicle body color identification. Correct rate.

The invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

As shown in Figure 1, the vehicle body color recognition method disclosed by the present invention comprises the following steps:

Step S110, acquiring the video of the vehicle on the road to be detected.

Shoot the road surveillance video in a static scene through the camera, obtain the position of the captured video frame in the video sequence, and read the video in milliseconds from the first frame; or directly import the existing road surveillance video, start from the first frame of the imported video Frame start, import the video frame by frame according to the time sequence of video frame playback.

Figure 2 is a video image of a vehicle to be identified. The white dotted line is the trip wire marked by the user, and the blue vehicle in contact with the trip wire is the vehicle to be recognized.

Step S120: Perform moving object detection on the video data, intercept the binarized foreground moving object circumscribed rectangular image P1 passing the tripwire, and then extract the color pixel value corresponding to the foreground moving object area in the rectangular image in the original video frame And get image P2.

According to their own needs, the user artificially draws a dotted line as a trip line at the start frame of the vehicle video (see Figure 2). Based on the VIBE (Visual Background Extractor) algorithm, the moving target in the foreground area is detected and obtained, and the binarized moving target image P1 is obtained (that is, the moving target pixel value in the image is 1, and the background pixel value is 0). The specific operation is that when it is detected that there is a moving target crossing the tripwire, that is, there is an intersection for the first time between the image foreground target pixel point set and the tripwire pixel point set on the image, then the circumscribed rectangle of the foreground moving target pixel point set ( That is, traverse the edge of the foreground moving object, record the pixel coordinates corresponding to the maximum value and the minimum value in the vertical and horizontal directions respectively, determine the circumscribed rectangle of the foreground moving object according to the calculated coordinates), and then intercept the circumscribed rectangle image to obtain binarization The vehicle image P1 of the rectangle, and then extract the color pixel value corresponding to the foreground moving target area in the rectangle (that is, the pixel area with a pixel value greater than 0) in the original video frame to obtain the vehicle image P2.

Step S130, calculate the maximum value Max(R, G, B) and the minimum value Min(R, G, B) of the RGB of each pixel in the image P2, and then calculate the difference between the maximum value and the minimum value to obtain the RGB color channel difference Image P3 (ie P3=Max(R,G,B)-Min(R,G,B)).

Step S140 , a threshold M1 is given, and the image P3 is thresholded to obtain a binarized image P4 (that is, the pixel value of the body area is 1, and the pixel value of the background area is 0). A pixel with a pixel value greater than M1 is 1, otherwise it is 0.

Step S150, counting the number of pixels N _P4 contained in the body area in the image P4 (that is, the area with a pixel value greater than 0), and counting the number N _P2 of all pixels in the circumscribed rectangular image P2 of the moving target in the foreground (that is, the length of the rectangular frame and The product of the number of pixels of the width of the rectangular frame), calculate the ratio of the number of pixels R=N _P4 /N _P2 .

Step S160, by judging the ratio R of the number of pixels, classify the vehicles into colored vehicles and black, white and gray vehicles. Given a threshold M2, if R is greater than the threshold M2, proceed to step S170 for color recognition; if R is smaller than the threshold M2, proceed to step S180 for black, white and gray recognition.

Step S170, define color value ranges in H (hue) space (see Table 1), which are defined as red, orange, yellow, green, cyan, blue, and purple, a total of 7 colors. Extract the color value corresponding to the area in the image P2 where the pixel value in the binarized image P4 is greater than 0, and then convert the RGB to the HSV color space (that is, composed of hue (Hue), saturation (Saturation) and brightness (Value) color model). Calculate the histogram of the H space, detect the color range of the H space where the highest peak in the histogram falls, and the color corresponding to the range is the color of the vehicle body.

Table 1: Define color value range in H (hue) space

Step S180, define color value ranges in the V (brightness) space (see Table 2), which are respectively defined as black, white, and gray in total. Convert the RGB color space of the foreground moving target P2 to the HSV color space, find the centroid of P2, and then draw a circle with the centroid as the center and the shortest distance from the centroid to the edge of the vehicle’s circumscribed rectangle as the radius. Divide the disk area into 5 sectors at 72 degrees, and calculate the histogram of the pixels in each sector area (excluding the pixels in the background area) in the V space, and detect the position of the highest peak in the histogram corresponding to the V space The color of the fan-shaped area is the color of the fan-shaped area. Finally, the color of the vehicle body is voted by voting, and the color with the largest number of fan-shaped areas is determined to be the body color of the vehicle.

Table 2: Define the color value range in V (brightness) space

Brightness space (0～255) black Ash White maximum value 0 47 221 minimum value 46 220 255

Step S190, storing the body color recognition result.

Step S200, judge whether it is the last frame of video, if not, go to step S120, continue to segment the foreground moving object and judge the color; if not, go to step S210.

Step S210, return all vehicle body colors. End the body color recognition process.

Fig. 3 is the histogram distribution of the color corresponding to the binarized body area in the image P3 in the H space. The histogram presents an obvious unimodal characteristic, so the present invention can better identify the color of the body.

The basic principles, main features and characteristics of the present invention have been shown and described above. Due to the use of the maximum value and minimum value difference algorithm of the color space, the influence of the vehicle shadow color and the window glass color on the body color recognition is reduced, and the accuracy of the body color recognition is improved. Secondly, the present invention can identify the color of the vehicle body from any angle, without extracting other features (such as using the license plate to locate the color region) to identify the color of the vehicle body.

Claims (8)

1. a kind of body color recognition methods, which is characterized in that comprise the following steps：

The video image and setting for obtaining vehicle are stumbled line, moving object detection are carried out to the vedio data of input, before interception The boundary rectangle of scape moving target；Then calculate in the boundary rectangle image in the RGB color passage of all pixels maximum with The difference of minimum value passes through a threshold value M1 so that the threshold value obtains binaryzation to RGB color error image into row threshold division Vehicle body image, to remove glass for vehicle window and vehicle shadow interference region；

Number of pixels ratio is calculated, distinguishes colored vehicle and black and white grey vehicle, the number of pixels ratio is foreground moving Target is by maxima and minima difference processing in RGB color passage and the number of pixels in the vehicle body region after binary conversion treatment The ratio between number of pixels included with the boundary rectangle of foreground moving object；

Using H spatial histograms red, orange, yellow, green, blue, blue, purple are amounted to seven kinds of colors to be identified；Using V spatial histograms Method is voted with the color based on sector region black, white, grey three kinds of colors altogether are identified；

Export the body color identified.

2. the method as described in claim 1, it is characterised in that：Maxima and minima difference in the RGB color passage Maximum Max (R, G, B) and minimum value of each pixel in rgb color space in image are identified to be used for body color The difference of Min (R, G, B), i.e. Max (R, G, B)-Min (R, G, B).

3. the method as described in claim 1, it is characterised in that：The number of pixels that the boundary rectangle of foreground target is included is to adopt With the moving target obtained in foreground area is detected based on VIBE (visualization background extracting) algorithm, when having detected moving target Across stumble line when, i.e., the line pixel point set of stumbling on display foreground object pixel point set and image has first time intersection, then calculates The number of pixels that the boundary rectangle of the foreground moving object pixel point set is included.

4. the method as described in claim 1, it is characterised in that：The H spatial histograms pass through rgb space for vehicle image After the maximum and minimum value of middle R, G and B triple channel subtract each other and carry out binary conversion treatment, obtained vehicle body region, i.e. pixel value Region more than 0, the H spatial histograms in corresponding color region in former video frame.

5. the method as described in claim 1, it is characterised in that：The color method of voting based on sector region is The RGB color of moving target external world rectangular image is converted into hsv color space, finds the barycenter of foreground moving object, Then using barycenter as the center of circle, drawn and justified as radius using the beeline of the barycenter to vehicle boundary rectangle frame edge, by disc area 5 sectors are divided by 72 degree, and calculate pixel in each sector region, the pixel without background area, the Nogata in V spaces Figure, it is the color of sector region to detect the position in histogram where top corresponding color, the color in V spaces, Finally possess the body color that the most color of sector region quantity is the vehicle.

6. a kind of body color recognition methods, which is characterized in that comprise the following steps：

Step S110. obtains road vehicle video to be detected；

Step S120. carries out moving object detection to video data, extracts the foreground moving object by line of stumbling, interception prospect fortune The boundary rectangle of moving-target；

Step S130. calculates the maximum Max (R, G, B) and minimum of the RGB color passage of each pixel in boundary rectangle image Value Min (R, G, B), the difference for then calculating maxima and minima obtain RGB color error image；

Step S140. gives a threshold value M1, and binaryzation is obtained into row threshold division to RGB color error image using the threshold value Image；

The number of pixels N in vehicle body region in step S150. statistics binary images P4_P4, count the external square of foreground moving object The number N of all pixels in image P2 corresponding to shape_P2, calculate number of pixels ratio R=N_P2/ N_P4；

Vehicle is divided into colored vehicle and black and white grey vehicle by step S160. by number of pixels fractional threshold determination methods；

Step S170. amounts to red, orange, yellow, green, blue, blue, purple using H spatial histograms seven kinds of face when vehicle is colored vehicle Color is identified；

Step S180. is when vehicle is achromaticity vehicle, using V spatial histograms and the color side of voting based on sector region Black, white, grey three kinds of colors altogether are identified in method；

Step S190. is stored or output recognition result.

7. body color recognition methods as claimed in claim 6, which is characterized in that H（Form and aspect）Color value defined in space Scope such as following table：

。

8. body color recognition methods as claimed in claim 6, which is characterized in that the V（Brightness）Color defined in space Value range such as following table：

Brightness space（0~255） It is black Ash In vain Maximum 0 47 221 Minimum value 46 220 255

。