KR100818317B1 - Method for recognizing a vehicle using the difference image - Google Patents

Abstract

The present invention relates to a method of recognizing a moving vehicle, and more particularly, to obtain a difference image through a subtraction operation on two images of a predetermined time interval taken by a camera, and to obtain the difference image. The present invention relates to a method of recognizing a moving vehicle by obtaining a candidate region by applying a window tracking method and then performing image processing on the candidate region.

According to an aspect of the present invention, there is provided a method for recognizing a moving vehicle, comprising: calculating a difference image from a target image to be recognized by a moving vehicle and a comparison image photographed immediately before the target image; It is assumed that a vehicle exists by moving a virtual window with respect to the calculated difference image and applying a window tracking method that calculates a sum of binary values of pixels for an area within the virtual window. Selecting a vehicle detection area to be used; Obtaining an edge image by edge transforming the target image; A pattern comparison step of comparing a shape pattern obtained in the vehicle detection area of the edge image obtained in the edge conversion step with a shape pattern of a model vehicle which is previously input; When it is determined that each pattern is similar to each other in the pattern comparison step, detecting the vehicle information for the preceding vehicle from the corresponding edge information; characterized in that it comprises a.

Vehicle Recognition, Difference Image, Window Tracking, Edge Filter

Description

Method for recognizing a vehicle using the difference image}

1 is a block diagram showing the configuration of a preceding vehicle collision warning device to which the present invention is applied.

2 is a flowchart illustrating a method for recognizing a moving vehicle using a vehicle image according to an embodiment of the present invention.

3 is a view illustrating a difference image calculating process of the difference image calculating step illustrated in FIG. 2.

FIG. 4 is a diagram illustrating an edge transformation process of the edge transformation step shown in FIG. 2. FIG.

FIG. 5 is a flowchart showing details of the vehicle detection region calculation step shown in FIG. 2; FIG.

<Explanation of symbols for the main parts of the drawings>

10: camera unit 20: image acquisition unit

30: image storage unit 40: central processing unit

50: alarm processing unit 60: alarm device

The present invention relates to a method of recognizing a moving vehicle, and more particularly, to obtain a difference image through a subtraction operation on two images of a predetermined time interval taken by a camera, and to obtain the difference image. The present invention relates to a method of recognizing a moving vehicle by obtaining a candidate region by applying a window tracking method and then performing image processing on the candidate region.

In general, in order to recognize an obstacle present in front of an object such as a vehicle and to take action, an image recognition device that recognizes information about the size and location of the obstacle by processing an image photographed by the camera. Is widely used.

The image recognition device may be applied to fields such as automobiles, robot technology (RT) or automated guided vehicles (AGV). In particular, in the automobile field, accurate vehicle recognition must be guaranteed even at high speeds. Is considered.

Advanced safety vehicle (ASV) related fields, which are currently being highly functionalized and intelligent in the automotive field, are the main applications of vehicle recognition devices, and the front obstacle warning system for preventing inadvertent driving. In addition, the market is expected to expand further in the future, as carmakers in each country are expanding their range of applications, such as rear parking guides to help beginners park, and brakes to prevent collisions in dangerous situations.

A conventional image recognition apparatus for recognizing a preceding vehicle in a moving vehicle acquires an image of a target to be recognized using a camera, and then extracts a feature of the target by using digital image processing technology, and uses the feature to extract the target. Perform the image processing process to check. Here, most conventional image recognition apparatuses recognize a preceding vehicle through a method of analyzing a pattern on edge information obtained by applying one image filter such as an edge filter to a captured target image. .

However, such a conventional vehicle recognition method recognizes a vehicle by performing only a simple image filtering operation on a single image acquired at a specific time, so that the acquired image is noisy, a building similar to a preceding vehicle, or the like. If it is included there is a problem that the recognition rate for the preceding vehicle is lowered.

The present invention is to solve the above problems of the vehicle recognition method according to the prior art. That is, an object of the present invention is to calculate a vehicle detection area in which a moving vehicle is likely to exist through a vehicle image in advance, and perform a pattern comparison on edge information of the vehicle detection area in an edge-converted image from a target image. The present invention provides a robust vehicle recognition method capable of improving the recognition rate of a moving vehicle and being less affected by noise components of a captured image.

The present invention as a technical idea for achieving the above object,

Calculating a difference image from a target image to recognize a moving vehicle and a comparison image photographed immediately before the target image; It is assumed that a vehicle exists by moving a virtual window with respect to the calculated difference image and applying a window tracking method that calculates a sum of binary values of pixels for an area within the virtual window. Selecting a vehicle detection area to be used; Obtaining an edge image by edge transforming the target image; A pattern comparison step of comparing a shape pattern obtained in the vehicle detection area of the edge image obtained in the edge conversion step with a shape pattern of a model vehicle which is previously input; In the pattern comparison step, if it is determined that each pattern is similar to each other, detecting the vehicle information for the preceding vehicle from the corresponding edge information; provides a mobile vehicle recognition method using a vehicle image, characterized in that it comprises a. .

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

First, prior to explaining a mobile vehicle recognition method using a vehicle image according to an embodiment of the present invention, a prior vehicle collision warning device to which the present invention is applied will be described.

1 is a block diagram showing the configuration of a preceding vehicle collision warning device to which the present invention is applied.

The preceding vehicle collision warning device to which the present invention is applied is a device mounted in a driving vehicle and warning the driver of a collision risk by recognizing a distance from the preceding vehicle. As shown in FIG. 1, a front image including a preceding vehicle is photographed. An image storing unit 10, an image obtaining unit 20 obtaining digital image data from an image captured by the camera unit 10, and an image storing image data received from the image obtaining unit 20. A storage unit 30, a central processing unit 40 for reading image data stored in the image storage unit 30 and recognizing a preceding vehicle from the read image data, an alarm device 60 for generating an alarm to the driver, and It comprises an alarm processing unit 50 for outputting the distance between the vehicle and the collision warning, etc. through the alarm device in accordance with the preceding vehicle information received from the central processing unit 40.

Here, the image acquisition unit 20 receives an analog image signal captured by the camera unit 10 and converts it into a digital image signal having a preset frame per second and resolution.

The central processing unit 40 reads the image data stored in the image storage unit 30, that is, the image frame, and performs image processing by applying a vehicle recognition algorithm, thereby recognizing a preceding vehicle located in the image frame. In addition, the central processing unit 40 calculates the distance between the preceding vehicle and the driving vehicle by using the position information of the vehicle among the information of the preceding vehicle, and thus determines whether there is a collision risk.

The present invention relates to a vehicle recognition method for recognizing a preceding vehicle from an image photographed in front of a driving vehicle by the central processing unit 40. Hereinafter, a method for recognizing a moving vehicle using a vehicle image according to an embodiment of the present invention. This will be described in detail with reference to the flowchart shown in FIG. 2.

First, a difference image is calculated from a target image and a comparison image photographed immediately before the target image (S110). That is, as shown in FIG. 3, the traveling vehicle is subtracted by subtracting the video image obtained at the previous time point T (k-1) from the video image at the specific time point T (k) to be recognized. And a difference image representing the difference between the two images generated due to the difference in speed of the preceding vehicle.

Thereafter, the vehicle detection area in which the vehicle is estimated to be present is selected by applying a window tracking method to the difference image calculated from the step S110 (S120). Here, the window tracking method is a method of calculating a feature of an image region within a virtual window while scanning a whole image region by moving a rectangular virtual window on the image image. Let's explain.

When the vehicle detection area calculated in the vehicle detection area calculation step S120 exists, an edge image is obtained by performing edge conversion on a target image (S130), and when the calculated vehicle detection area does not exist, the vehicle is detected. Terminate the entire recognition process.

In the edge conversion step (S130), the edge filter is applied to a target image composed of color color information as shown in FIG. 4A, and is converted into an edge image composed of binary color information as illustrated in FIG. 4B. As the edge filter, it is preferable to use a Sobel Edge Filter having excellent edge detection performance. When using a Sobel edge filter, the filter threshold value is preset, and a point having a brightness value smaller than the threshold value is converted to a value of 0, and a point having a brightness value larger than that value is converted to a value of 1, so that Binary edge information is obtained. The threshold is preferably set to 50 to 60.

When the edge conversion step (S130) is completed, a pattern comparison step (S140) is performed in which the shape pattern obtained from the horizontal and vertical edge information in the vehicle detection area of the edge image and the shape pattern of a model vehicle that are inputted are compared with each other. . The shape pattern may use a pattern using various geometric information. For example, a model vehicle may be located in a corresponding vehicle detection area by using a calculation of a width of vehicle on a horizon according to a perspective such as Equation 1 below. By comparing the calculated vehicle width and the width obtained from the edge information with each other, it is possible to determine whether the shape patterns are similar.

Vehicle width = ((col / a) * F * secθ * X) / (z-tanθ)

(here,

col: the vertical resolution of the camera (pixels),

a is the horizontal axis length of the CCD cell,

F: focal length of the lens,

θ: the angle of the camera to the horizon,

X: model vehicle width (e.g. regular vehicle = 2.0m),

z: Distance of camera below the vehicle detection area (m))

Here, in addition to the vehicle width described above, the vehicle height may be calculated and used for pattern comparison in a similar manner. In addition, since the geometric pattern of the model vehicle can be changed according to the vehicle type, it is desirable to improve the accuracy by comparing with the shape patterns of the various vehicle types as possible.

When it is determined that the patterns are similar in the pattern comparison step (S140), vehicle information such as the position and size of the preceding vehicle is detected from the corresponding edge information (S150), and when it is determined that they are not similar, the vehicle in the target image It is determined that it does not exist and the vehicle recognition process ends.

Finally, when it is determined that the edge image of the pattern similar to the model vehicle, the corresponding edge image is recognized as a vehicle and detected (S150).

On the other hand, in the edge conversion step (S130), instead of edge conversion of the entire target image, edge conversion of only the vehicle detection area obtained in the vehicle detection area calculation step (S120) may reduce the calculation time required for edge conversion. have.

Hereinafter, a two-dimensional coordinate system in which the lower left side of the video image is the origin, the horizontal direction is the x-axis, and the vertical direction is the y-axis is set, and the vehicle detection area calculating step S120 described above with reference to FIG. Let's do it.

First, the difference image obtained in the difference image calculation step S110 is binarized, and the initial size of the virtual window is determined in consideration of the focal length and mounting angle of the camera and the width of the actual vehicle. Here, the binarized difference image data may be obtained by binarizing brightness values of respective pixels of the difference image based on a specific threshold value.

Thereafter, the lower left corner of the virtual window is positioned at the origin of the difference image, the virtual window is moved at regular intervals, and the window calculates an image sum obtained by summing all binary values of each pixel in the virtual window region at each position. Perform a trace. Here, as shown in (a) of FIG. 5, the virtual window is moved by a predetermined interval Δx along the x-coordinate direction, and when the right end of the difference image is reached, as shown in (b) of FIG. 5, it is fixed in the y-coordinate direction. While moving by the interval Δx, the virtual window is reduced in proportion to the perspective, and the virtual window is moved from the initial x coordinate position to the left to right again.

When the window tracking for the entire area of the difference image is completed, the position having the maximum value is determined by comparing the sum of the images with the width of the virtual window at each position where the virtual window is moved, and the corresponding position as shown in (c) of FIG. 5. Sets the virtual window area to the vehicle detection area. Here, the area is set as the vehicle detection area only when the maximum value of the image sum to the width of the virtual window, which is finally calculated, becomes equal to or greater than a predetermined value.

The present invention described above is not limited to the above-described embodiments and the accompanying drawings, and various substitutions, modifications, and changes are possible in the art without departing from the technical spirit of the present invention. It will be clear to those of ordinary knowledge.

As described above, in the mobile vehicle recognition method using the vehicle image according to the present invention, the vehicle detection region having a high possibility that the mobile vehicle exists in the vehicle image is calculated in advance, and the vehicle detection region is selected from the edge-converted images from the target image. Since the pattern comparison is performed on the edge information of the target image, even if the target image includes various objects or noise components including buildings, etc., in addition to the moving vehicle, the target image is simply compared with a method of recognizing the target image using a single image filter. There is an effect that can obtain a better recognition rate.

Claims (2)

In the method for recognizing a moving vehicle from the road image captured by the camera, Calculating a difference image from a target image to recognize a moving vehicle and a comparison image photographed immediately before the target image; It is assumed that a vehicle exists by moving a virtual window with respect to the calculated difference image and applying a window tracking method that calculates a sum of binary values of pixels for an area within the virtual window. Selecting a vehicle detection area to be used; Obtaining an edge image by edge transforming the target image; A pattern comparison step of comparing a shape pattern obtained in the vehicle detection area of the edge image obtained in the edge conversion step with a shape pattern of a model vehicle which is previously input; If it is determined that the patterns are similar to each other in the pattern comparison step, detecting vehicle information on the preceding vehicle from corresponding edge information; Moving vehicle recognition method using a vehicle image, characterized in that comprises a. The method of claim 1, The vehicle detection region selection step, Binarizing the difference image obtained in the difference image calculating step, and determining an initial size of the virtual window in consideration of the focal length and mounting angle of the camera and the width of the actual vehicle; Locate the lower left corner of the virtual window at the origin of the difference image, move the virtual window at regular intervals, and calculate a window sum that calculates the sum of all the binary values for each pixel inside the virtual window area at each location. Performing; When the window tracking step is completed, a position having a maximum value is determined by comparing the sum of images of the corresponding virtual windows at each position where the virtual window is moved, and setting the virtual window region at the corresponding position as the vehicle detection region. Doing; Consisting of The window tracking step, Moving vehicle recognition method using a vehicle image, characterized in that the size of the virtual window is reduced by a certain ratio as the virtual window is moved in the vertical direction.

Images