KR101276073B1 - System and method for detecting distance between forward vehicle using image in navigation for vehicle - Google Patents

Abstract

Disclosed are a distance recognition system and method through front vehicle detection in an image in a vehicle navigation. An image processor which receives an image of the front of the vehicle from the camera and converts the image into a gray image; A region setting unit for setting a region of interest for detecting the front vehicle in the gray image; A vehicle detector detecting a front vehicle in the ROI by using learning data about the vehicle; And a distance determination unit determining a distance from the vehicle in front of the vehicle based on the setting value of the distance in the image.

Description

System and method for distance recognition by detecting front car in image in vehicle navigation {SYSTEM AND METHOD FOR DETECTING DISTANCE BETWEEN FORWARD VEHICLE USING IMAGE IN NAVIGATION FOR VEHICLE}

Embodiments of the present invention relate to a system and method for detecting a distance from a front vehicle by detecting a front vehicle in a camera-integrated navigation.

In general, a vehicle is provided with a room mirror and a side mirror so that the driver can locate the surrounding objects, especially the vehicles, while driving. Normally, a room mirror is used to identify objects in the rear, and a side mirror is used to identify objects in the rear left and right.

However, there are blind spots in which objects cannot be identified by room mirrors and side mirrors, which causes a lot of traffic accidents. In addition, there is a problem that it is difficult to identify the left, right and rear objects by the room mirror and the side mirror in the case of snow or rain.

In order to solve this problem, vehicle collision warning devices have been developed and released for recognizing a vehicle in front of the vehicle to give a front collision warning. Such a vehicle collision warning device is equipped with a dedicated camera on the vehicle windshield behind the room mirror to capture the front image with the camera, recognize the front vehicle from the captured image, and detect an alarm when adjacent vehicles within a certain distance are detected. To let the driver know that there is a possibility of a crash.

Existing vehicle collision warning device is composed of a separate product equipped with a dedicated camera module, if necessary, the individual has to install separately in the vehicle. In particular, since accurate numerical data on the stereo camera or the camera installation location is required to recognize the distance in the image of the vehicle collision warning device, the camera installation has been difficult because of the accuracy required for installing the camera. In addition, although the use of navigation has become common recently, there has been pointed out that there is no navigation for providing a collision warning service by recognizing the distance between vehicles in front of the vehicle through the image.

Rather than providing a vehicle collision warning service using a camera-specific module, a distance recognition system and method are provided that can detect a distance from a front vehicle by directly detecting a front vehicle from an image in a camera-integrated navigation system.

Rather than using numerical data for a camera or camera installation position, a front vehicle distance recognition system and method are provided that can detect distance to a front vehicle by applying a threshold value obtained by an experiment.

There is provided a distance recognition system and method for recognizing a distance through front vehicle detection using only an image in a navigation without using a separate sensor to detect a front vehicle.

An image processor which receives an image of the front of the vehicle from the camera and converts the image into a gray image; A region setting unit for setting a region of interest for detecting the front vehicle in the gray image; A vehicle detector detecting a front vehicle in the ROI by using learning data about the vehicle; And a distance determination unit determining a distance from the vehicle in front of the vehicle based on the setting value of the distance in the image.

According to one side, the learning data may refer to data learned through an Adaboost algorithm for the image of the vehicle.

According to another aspect, the vehicle detector may detect at least one vehicle candidate region in the ROI, and then detect the front vehicle in the vehicle candidate region using a cumulative value corresponding to time of the vehicle candidate region.

According to another aspect, the vehicle detector may detect at least one vehicle candidate region in the ROI, and then detect the front vehicle in the vehicle candidate region using the size ratio of the vehicle candidate region.

According to another aspect, the vehicle detector detects a lane in the ROI and then detects the front vehicle according to the size ratio of the vehicle candidate area based on the lane.

According to another aspect, the vehicle detector converts a gray image corresponding to an ROI into an edge image from which an edge is detected, and then applies a hough transform algorithm to the edge image to determine a straight line of the edge. The location can be detected by lane.

According to another aspect, the vehicle detector detects at least one vehicle candidate region in the ROI, and then determines a vehicle ahead of the vehicle candidate region according to a detection frequency at which the vehicle candidate region is detected as a front vehicle by comparing with a previous image. Can be.

According to another aspect, the setting value for the distance may mean a value measured by using the size of the image that the actual vehicle forms on the image captured by the camera by the focal length of the camera.

According to another aspect, the setting value for the distance is a table value by matching the actual distance by the position of each area of the distance display grid using a distance display grid that divides the image taken by the camera into the same size area. Can mean.

An image processing step of receiving an image photographing the front of the vehicle from a camera and converting the image into a gray image; A region setting step of setting a region of interest for detecting a front vehicle in a gray image; A vehicle detecting step of detecting a front vehicle in the ROI by using learning data about the vehicle; And a distance determining step of determining a distance from the vehicle in front of the vehicle based on a setting value for the distance in the image.

By directly detecting the front vehicle and recognizing the distance from the front vehicle in the camera-integrated navigation system, the vehicle navigation warning function can be performed on its own without using a separate vehicle collision warning device, thereby enhancing the product competitiveness of the navigation in terms of functionality. Can be.

It is possible not only to detect the front vehicle by using the image from the navigation without using a separate sensor, but also to detect the distance from the front vehicle by applying the threshold value obtained by experiment without using numerical data on the camera or the camera installation position. This enables navigation products with vehicle crash warning at minimal cost.

1 is a block diagram illustrating an internal configuration of a system for recognizing a vehicle distance in front of a vehicle navigation system according to an embodiment of the present invention.
2 illustrates an example image for explaining a process of detecting a vehicle candidate region according to an embodiment of the present invention.
3 to 5 illustrate example images for explaining a process of determining a front vehicle in a vehicle candidate region according to an embodiment of the present invention.
FIG. 6 illustrates an example image for explaining a process of determining a distance from a vehicle in front of an embodiment of the present invention.
7 is a flowchart illustrating an entire process of a method for recognizing a vehicle distance in front of a vehicle according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an internal configuration of a system for recognizing a vehicle distance in front of a vehicle navigation system according to an embodiment of the present invention. The front vehicle distance recognizing system 100 is applied to a vehicle navigation device, in particular, a camera-integrated navigation device, and consists of one configuration with the navigation device.

The front vehicle distance recognition system 100 according to an embodiment is configured to detect a front vehicle by using computer vision technology in the navigation and recognize the distance to the front vehicle in advance to warn of the danger of collision with the front vehicle. As shown in FIG. 1, the camera 110 includes an image processor 120, an area setting unit 130, a vehicle detector 140, and a distance determiner 150. Here, the front vehicle distance recognition system 100 may recognize the distance between the front vehicle and the front vehicle in the image in the navigation by using the open computer vision algorithm openCV code.

The camera 110 is a means for photographing the front of the driving vehicle, and may photograph an image of a predetermined size (for example, 640 * 480). In this case, the captured image may refer to a red green blue (RGB) image. The camera 110 is integrated in a vehicle navigation system, and is mounted at a position facing the windshield of the vehicle, which is behind the navigation body so that the front of the vehicle can be photographed.

The image processor 120 receives an RGB image captured by the camera 110 and converts the input RGB image into a gray (black and white) image. In addition, the image processor 120 may perform smoothing on the contrast value of the gray image so that the gray image has a constant intensity value distribution.

If a part of the road is affected by the shadow, the vehicle may incorrectly recognize the distance between the vehicle in front of the vehicle or the vehicle in front of the vehicle. In order to minimize the impact of the shadow, the present embodiment may apply a method of correcting the light source. For example, the camera 110 may correct the light source to be used for photographing, and then photograph the front of the vehicle with the corrected light source to obtain an RGB image. As another example, the image processor 120 may apply a light source correction algorithm to the RGB image to correct the light source of the RGB image, and then change the corrected RGB image to a gray image. On the other hand, due to the influence of the subject reflected on the glass inside the vehicle may fail to detect the front vehicle in a specific portion of the image may not recognize the distance between vehicles. In order to improve this problem and minimize the effects of reflection, a corrected image can be obtained by attaching a reflection suppression box to the front end of the lens of the camera 110.

The region setting unit 130 sets a region of interest (ROI) for detecting the front vehicle in the gray image. In this case, the ROI may be set to an area in which the vehicle may exist in the captured image according to the installation angle and the viewing angle of the camera 110. For example, the area setting unit 130 may set an area of interest as much as a predetermined distance in the y-axis direction from the position where the lane starts. In this case, the position at which the lane starts, which is a condition for determining the ROI, and the predetermined distance may be estimated based on the width of the lane and the viewing angle of the camera 110.

The vehicle detector 140 detects the vehicle ahead of the vehicle using the learning data about the vehicle. In order to generate a learning pattern for detecting the front vehicle, a positive image is collected from the image captured by the camera as a positive image, and a non-vehicle-related image is collected as a negative image. Can be. In this embodiment, the Adaboost algorithm is applied to train training data classified into a positive image and a negative image, and the trained learning result is changed into an XML file, which is then used as a vehicle detection pattern for detecting a front vehicle. It is available. That is, as shown in FIG. 2, the vehicle detector 140 detects at least one vehicle candidate region 201, 202 by applying the Adaboost algorithm in the region of interest, and then selects the vehicle candidate region 201, 202. ), The vehicle in front of the vehicle can be detected. For example, the vehicle detection unit 140 may verify the vehicle candidate region based on the fact that the vehicle becomes brighter when a cumulative contrast value is accumulated over time in the image. As shown in FIG. 3, the vehicle detection unit 140 calculates a light and shade cumulative value over time for each vehicle candidate region, and determines that the vehicle candidate region 301 whose light and light cumulative value exceeds a predetermined threshold value is a front vehicle. If the area is not, it may be excluded from the vehicle candidate area. As another example, the vehicle detector 140 may verify the vehicle candidate region using the distance ratio with the front vehicle in the spatial region of one image. Referring to FIG. 4, in the case of a front vehicle, a vehicle using a spatial ratio characteristic, which is a characteristic in which the left and right width of the vehicle 401 decreases as the distance between the vehicles increases, and the left and right width of the vehicle 402 widens as the distance between the vehicles decreases. The experimental value of the distance ratio according to the position can be obtained. That is, the vehicle detection unit 140 calculates the size ratio for each vehicle candidate region, and then determines that the vehicle candidate region is the front vehicle when the size ratio of the vehicle candidate region corresponds to the experimental value of the distance ratio according to the vehicle position. Can be excluded from the candidate area. As another example, the vehicle detector 140 may detect a lane in the ROI, and then verify the vehicle candidate area through a distance ratio with the front vehicle based on the detected lane. To this end, the vehicle detector 140 may obtain an edge image of detecting an edge by applying a Canny algorithm to a gray image corresponding to the ROI. As illustrated in FIG. 5, the vehicle detector 140 may apply a Hough transform algorithm to the edge image to detect an edge representing a straight line and recognize the detected position of the straight line as the lane 503. At this time, the vehicle detection unit 140 selects the vehicle candidate region at the position of the lane 503 among the vehicle candidate regions, and if the size ratio of the selected vehicle candidate region corresponds to the experimental value of the distance ratio according to the vehicle position, the vehicle ahead of the vehicle (501) can be determined. When the above lane detection algorithm is applied to detect the front vehicle, the front vehicle detection using the distance ratio with the front vehicle can be more accurate. As another example, the vehicle detector 140 may detect the front vehicle by applying the vehicle detection frequency of the previous images and the current image at the position detected as the vehicle candidate region. In other words, the vehicle detector 140 may compare the vehicle candidate region with the previous image to determine whether the vehicle candidate region is the front vehicle according to the detection frequency at which the vehicle candidate region is detected as the front vehicle.

The distance determiner 150 determines a distance to the vehicle ahead based on a setting value (hereinafter, referred to as a 'distance experiment value') for the distance in the image obtained through the experiment. For example, the distance determiner 150 may recognize the distance to the front vehicle by using the distance test value measured based on an internal parameter of the camera 110. This plays an important role in identifying the relationship between the real three-dimensional world and the image acquired by the camera 110. This is also important in the unit of the real world and the unit relationship in the image. The camera calibration process informs the geometric model of the camera and the distortion model of the lens. These two models can be used to define the internal parameters of the camera and to interpret the actual scene. The distance can be measured by calculating how much the actual object forms in the image by the focal length of the camera. That is, the experiment can obtain a distance experiment value by using the size of the image formed on the image captured by the actual vehicle by the focal length of the camera, the distance determiner 150 is based on the internal parameters of the camera 110 The distance from the front vehicle detected by the vehicle detector 140 may be determined to be far or near based on the measured distance test value. As another example, the distance determiner 150 may recognize the distance to the front vehicle by using an experimental value obtained through experiments without using pure camera measurement parameters. Camera parameters are not used in the process of finding the unit relationship between the real world and the image. In the real world, a distance table such as a grid and a ruler may be projected by a camera to define a distance table using the projected image. As shown in Fig. 6, although the grids are the same size, their distance values are different from those of the real world depending on their positions. However, since it has a certain distance according to the grid position, the distance can be determined from the image by making the distance table. That is, the distance experiment value may be tabulated by matching the actual distance for each position of each area of the distance display grid using a distance display grid that divides the image photographed by the camera 110 into regions having the same size through experiments. The distance determiner 150 may determine whether the distance to the front vehicle detected by the vehicle detector 140 is far or near based on the distance experiment value of the predefined distance table. At this time, the distance determination unit 150 may directly provide the distance to the vehicle in front of the numerical value because the actual distance value is matched for each region position.

The front vehicle distance recognizing system 100 having the above-described configuration recognizes the front vehicle in the image of the navigation and generates an alarm sound when it is determined that the distance to the front vehicle is close to or within a predetermined distance, so that the driver may contact the front vehicle with the front vehicle. The risk of collision can be warned in advance.

7 is a flowchart illustrating an entire process of a method for recognizing a vehicle distance in front of a vehicle according to an embodiment of the present invention. In the method of recognizing the front vehicle distance according to the present embodiment, respective steps may be performed by the front vehicle distance recognition system 100 of the vehicle navigation described with reference to FIG. 1.

In operation 710, the front vehicle distance recognition system 100 receives an RGB image captured by a camera photographing the front of the vehicle, and converts the RGB image into a gray image. In this case, the front vehicle distance recognition system 100 may perform smoothing on the contrast value of the gray image so that the gray image has a constant contrast value distribution, and may use a light source used in the camera to minimize the shadow effect on the image. After calibrating or calibrating the light source of the RGB image, which is a captured image, the gray image can be changed.

In operation 720, the front vehicle distance recognition system 100 sets a region of interest for detecting the front vehicle in the gray image. In this case, the ROI may be set to an area in which the vehicle may exist in the captured image according to the installation angle and the viewing angle of the camera 110.

In operation 730, the front vehicle distance recognizing system 100 may detect the front vehicle in the ROI using learning data of the vehicle learned by the Adboost algorithm. For example, the front vehicle distance recognizing system 100 may detect at least one vehicle candidate region in the ROI, and then detect the front vehicle using a cumulative value corresponding to time of the vehicle candidate region. As another example, the front vehicle distance recognition system 100 detects at least one vehicle candidate region in the ROI, and then detects the front vehicle according to the size ratio of the vehicle candidate region based on an experimental value of the distance ratio according to the vehicle position. can do. At this time, the front vehicle distance recognition system 100 detects a lane in the ROI, selects a vehicle candidate region at a position of the lane among the vehicle candidate regions, and then selects a ratio of the size of the selected vehicle candidate region according to the vehicle position. The front vehicle can be recognized by comparing with the experimental value of. As another example, the front vehicle distance recognition system 100 detects at least one vehicle candidate region in the ROI, and then applies the vehicle detection frequency of the previous images and the current image at the position detected as the vehicle candidate region in front of the vehicle. The vehicle can be detected.

In step 740, the front vehicle distance recognition system 100 may determine the distance to the front vehicle based on the distance experiment value in the image obtained through the experiment. Here, as a distance experiment value for determining the distance to the front vehicle, a value measured by using the size of the image formed on the image captured by the actual vehicle by the camera by the focal length of the camera may be used. In addition, using the distance display grid that divides the image captured by the camera into regions of the same size, the distance table tabled by matching the actual distance values for each position of the distance display grid can be used as the distance experiment value in the image. have.

As described above, according to embodiments of the present disclosure, a camera-integrated navigation device having a vehicle collision warning function may be provided through a configuration that directly performs front vehicle detection and distance recognition with a front vehicle. In addition, according to embodiments of the present invention, the distance to the front vehicle may be detected using a distance test value in an image obtained through experiments without using a separate sensor or pure camera measurement parameter.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above embodiments, and those skilled in the art to which the present invention pertains various modifications and variations from such descriptions. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the equivalents of the claims, as well as the claims.

100: front vehicle distance recognition system
110: camera
120: image processing unit
130: area setting unit
140: vehicle detection unit
150: distance determination unit

Claims (21)

In the vehicle distance recognition system of the front of the vehicle navigation,
An image processor which receives an image of the front of the vehicle from a camera and converts the image into a gray image;
A region setting unit for setting a region of interest for detecting a front vehicle in the gray image;
A vehicle detector detecting the front vehicle in the ROI by using learning data about the vehicle; And
Distance determination unit for determining the distance to the front vehicle on the basis of the set value for the distance in the image
Lt; / RTI >
The vehicle detection unit,
Detecting at least one vehicle candidate region in the ROI, and detecting the front vehicle in the vehicle candidate region by using a cumulative contrast value over time of the vehicle candidate region;
Front vehicle distance recognition system of the vehicle navigation, characterized in that. The method of claim 1,
Wherein the image processing unit comprises:
Performing smoothing on the contrast value of the gray image
Front vehicle distance recognition system of the vehicle navigation, characterized in that. The method of claim 1,
The learning data,
Data learned through an Adaboost algorithm for the image of the vehicle
Front vehicle distance recognition system of the vehicle navigation, characterized in that. delete delete delete delete delete The method of claim 1,
The setting value for the distance is,
It is a value measured using the size of an image formed by an actual vehicle on an image captured by the camera by the focal length of the camera.
Front vehicle distance recognition system of the vehicle navigation, characterized in that. The method of claim 1,
The setting value for the distance is,
It is a table value by matching the actual distance for each position of each area of the distance display grid by using a distance display grid for dividing the image taken by the camera into an area of the same size
Front vehicle distance recognition system of the vehicle navigation, characterized in that. In the vehicle vehicle distance recognition method of the vehicle navigation,
An image processing step of receiving an image photographing the front of the vehicle from a camera and converting the image into a gray image;
A region setting step of setting a region of interest for detecting a front vehicle in the gray image;
A vehicle detecting step of detecting the front vehicle in the region of interest using learning data about the vehicle; And
A distance determination step of determining the distance to the front vehicle on the basis of the set value for the distance in the image.
Lt; / RTI >
The vehicle detection step,
Detecting at least one vehicle candidate region in the ROI, and detecting the front vehicle in the vehicle candidate region by using a cumulative contrast value over time of the vehicle candidate region;
A method for recognizing a vehicle distance ahead of a vehicle navigation, characterized in that. The method of claim 11,
The image processing step,
Performing smoothing on the contrast value of the gray image
A method for recognizing a vehicle distance ahead of a vehicle navigation, characterized in that. The method of claim 11,
The learning data,
Data learned through an Adaboost algorithm for the image of the vehicle
A method for recognizing a vehicle distance ahead of a vehicle navigation, characterized in that. delete delete delete delete delete The method of claim 11,
The setting value for the distance is,
It is a value measured using the size of an image formed by an actual vehicle on an image captured by the camera by the focal length of the camera.
A method for recognizing a vehicle distance ahead of a vehicle navigation, characterized in that. The method of claim 11,
The setting value for the distance is,
It is a table value by matching the actual distance for each position of each area of the distance display grid by using a distance display grid for dividing the image taken by the camera into an area of the same size
A method for recognizing a vehicle distance ahead of a vehicle navigation, characterized in that. A computer-determinable recording medium having recorded thereon a program for executing the method of any one of claims 11 to 13, 19 and 20.

Images