JP2000113169A - Device and method for detecting vehicle - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a vehicle detector for exactly detecting a vehicle in the object area of an image while using the lower shadow area of a car body concerning a device for detecting the vehicle from the video of a monitor camera installed on a road through image processing. SOLUTION: A road surface area and a dark area are detected by a road surface area detecting means 13 and a dark area detecting means 15 the timewise and spatial relation of detected road surface area and dark area is collated by a lower shadow area means 16, and only when the road surface area is connected at the back of the dark area, the dark area is discriminated as the lower shadow area of an object so that only the true lower shadow area can be properly extracted without erroneously extracting the dark area of the car body.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing technique for a traffic flow monitoring system that automates a monitoring camera installed on a road by image processing, performs automatic measurement of a traffic flow, and automatically detects an unexpected event. The present invention relates to a vehicle detection device and a vehicle detection method for immediately and reliably detecting a vehicle entering a monitoring area in an image.

[0002]

2. Description of the Related Art In recent years, there has been a demand for the development of a monitoring system that automates a surveillance camera installed on a highway, a highway, or the like by image processing to automatically measure a traffic flow or automatically detect an unexpected event. .

The following has been proposed as a technique for detecting a vehicle that has entered an image.

Reference 1: Takahashi, Kitamura, Sato: JP-A-10-
No. 154292, "Traffic flow monitoring device", Document 2; Segawa: JP-A-8-147475, "Moving object detection device", Document 3; Hayashi: JP-A-9-81889, "Vehicle detection device", Document 4; Enomoto, Arasawa: "Image-based traffic flow measurement method and apparatus" in JP-A-7-210799.

However, these methods can be applied only in a limited range of installation environment and installation angle of view.

For example, considering the image of a surveillance camera inside a tunnel, both the vehicle and the background are generally dark and the contrast is not clear. That is, since the pattern and the edge of the vehicle body are not clear, the method of detecting the vehicle edge and detecting the vehicle as described in References 1, 2, and 3 cannot be applied.

Further, the installation height of the surveillance camera inside the tunnel is low, and even if the camera is installed on the ceiling of the tunnel, the installation height is only several tens of cm above the ceiling surface of a large vehicle such as a truck or a bus. In such a low camera position, when a large vehicle passes through a lane close to the camera, the entire screen is occupied by the body of the large vehicle. In the background difference method used in Reference 4, the entire screen is extracted as a change area, and meaningful data cannot be extracted in subsequent processing. That is, in the background subtraction method, even if feature points are extracted or the vehicle space occupancy is calculated from the difference area, the data does not become meaningful data.

[0008] Further, in all of Document 1, Document 2, Document 3, and Document 4, since the assumed camera position is high, it is assumed that the traveling lane of the vehicle can be determined from the vehicle detection position in the image. However, in the case of the above-described example of a large vehicle photographed by the tunnel monitoring camera, it is difficult to determine the actual traveling position from the position of the vehicle body region in the image or the position of a characteristic point of the vehicle body. The position in the space cannot be estimated unless it is possible to correctly determine where the position of the feature point corresponds on the vehicle body.

Therefore, there is a need for a method of detecting a vehicle based on features that can be extracted from a low angle of view, such as a tunnel surveillance camera, can be extracted from all types of vehicles, and have a clear feature in the vehicle body. . A feature that satisfies such a condition is the shadow area of the vehicle body. In addition,
The “lower shadow area” is a shadow of the vehicle body reflected on a road surface located near and below the vehicle body.

[0010] Assuming that the surveillance camera takes an image of the vehicle from the rear in the traveling direction, there is only one shadow area on the road surface behind all vehicles regardless of the type of vehicle. Since the position of the lower shadow area is determined on the road surface, it is possible to determine the traveling lane.

There is a method of extracting an area having a luminance lower than a predetermined threshold value in order to detect a lower shadow area. However, depending on the type of vehicle, there may be a region having a dark luminance in the vehicle body as in the shadow region. For example, a hood of a truck, a black window frame of a bus, and a rear window of a van are also extracted as dark areas depending on lighting conditions.

[0012]

As described above, in order to detect a vehicle from an image photographed in a dark environment such as a surveillance camera inside a tunnel, the lower shadow area of the vehicle body is detected. However, there is a problem that if a low-luminance region is extracted only by a single threshold value, a dark region of the vehicle body is erroneously extracted as a shadow.

Accordingly, the present invention provides a vehicle detection apparatus and method for accurately detecting a vehicle in a target area in an image using an under shadow area of a vehicle body.

[0014]

According to a first aspect of the present invention, there is provided a vehicle detection apparatus for detecting a vehicle in a target area in an image based on an image of a monitoring target range in which the vehicle travels on a road surface. A pixel in a predetermined first luminance range in the target region of the image, a road surface region detection unit that detects the pixel as a road surface region, and a pixel in a predetermined second luminance range in the target region of the image. Pixel, a dark area detecting unit that detects the dark area, a road surface area detected by the road area detecting unit, and a temporal relationship or a spatial relationship between the dark area detected by the dark area detecting unit. Are compared, and as a result of the comparison, temporally behind the dark area, or
Only when the road surface area is continuously connected spatially rearward, the dark area is determined to be the lower shadow area of the vehicle, and the lower shadow determination unit that detects the vehicle based on the dark area is determined. It is a vehicle detection device.

According to a second aspect of the present invention, there is provided the vehicle detecting apparatus according to the first aspect, wherein the image is a spatio-temporal image in which certain areas are stacked with time.

According to a third aspect of the present invention, there is provided a vehicle detecting device for detecting the vehicle in a target area in the image based on an image of a monitoring target range in which the vehicle travels on a road surface. A vehicle intrusion detection area is set, a change pattern extraction unit that extracts a temporal change pattern of the luminance of the intrusion detection area, and a luminance level of the change pattern extracted by the change pattern extraction unit,
A luminance level classification unit that classifies into at least three classes of dark luminance, road surface luminance, and other luminance, and the change pattern classified by the luminance level classification unit continues the dark luminance level for a predetermined duration or longer. And a vehicle intrusion detection unit that detects that the vehicle has entered when the brightness level subsequently changes to a road surface luminance level.

According to a fourth aspect of the present invention, there is provided a vehicle detection device for detecting the vehicle in a target area in the image based on an image of a monitoring target range in which the vehicle travels on a road surface. A vehicle intrusion detection area is set, a spatiotemporal image is generated by stacking the intrusion detection area with time, and a spatiotemporal image generating unit, and the luminance level of the spatiotemporal image generated by the spatiotemporal image generating unit is A luminance level classifying unit for classifying into at least three classes of dark luminance, road surface luminance, and other luminance, and a dark luminance level region of the spatiotemporal image classified by the luminance level classifying unit is longer than a predetermined duration. Continue, and then if the road surface brightness level area connects in the direction immediately after in time,
A vehicle detection device, comprising: a vehicle intrusion detection unit that detects that a vehicle has entered.

According to a fifth aspect of the present invention, there is provided a vehicle intrusion verification unit for verifying the intrusion of the vehicle based on an image at a time when the vehicle intrusion detection unit detects the intrusion of the vehicle. A vehicle detection device according to claim 3 or 4.

According to a sixth aspect of the present invention, there is provided a vehicle detection method for detecting the vehicle in a target area in the image based on an image of a monitoring target range in which the vehicle travels on a road surface. A pixel in a predetermined first luminance range in the road area detection step of detecting as a road area, and a pixel in a predetermined second luminance range in the target area of the image, a dark area A temporal area or a spatial relation between a road area detected by the road area detecting step and a dark area detected by the dark area detecting step, and the matching is performed. As a result, temporally behind the dark area,
Alternatively, only when the road surface area is continuously connected to the spatial rear, the dark area is determined to be the shadow area of the vehicle, and the shadow determination step of detecting the vehicle based on the dark area is performed. This is a vehicle detection method.

The invention according to claim 7 is the vehicle detection method according to claim 6, wherein the image is a spatiotemporal image in which certain areas are stacked with time.

According to an eighth aspect of the present invention, there is provided a vehicle detection method for detecting the vehicle in a target area in the image based on an image of a monitoring target range in which the vehicle travels on a road surface. A change pattern extraction step of setting a vehicle intrusion detection area, and extracting a temporal change pattern of the luminance of the intrusion detection area; and a luminance level of the change pattern extracted by the change pattern extraction step is set to a dark luminance, a road surface. A luminance level discriminating step for discriminating into at least three classes of luminance and other luminance, and the change pattern separated by the luminance level discriminating step continues the dark luminance level for a predetermined duration or more, When the brightness level changes,
A vehicle detection method comprising: a vehicle intrusion detection step of detecting that a vehicle has entered.

According to a ninth aspect of the present invention, there is provided a vehicle detection method for detecting the vehicle in a target area in the image based on an image of a monitoring target range in which the vehicle travels on a road surface. A vehicle intrusion detection area is set, a spatiotemporal image is generated by stacking the intrusion detection area with time progress, and a spatiotemporal image generating step, and the luminance level of the spatiotemporal image generated by the spatiotemporal image generating step is A luminance level classifying step of classifying into at least three classes of dark luminance, road surface luminance, and other luminances; and a dark luminance level region of the spatiotemporal image classified by the luminance level separating step is longer than a predetermined duration. The vehicle intrusion detecting step for detecting that the vehicle has entered when the area of the road surface luminance level is connected in the direction immediately after the time. A vehicle detection method characterized by comprising more.

According to a tenth aspect of the present invention, there is provided a vehicle intrusion verification step for verifying the intrusion of the vehicle based on an image at a time when the vehicle intrusion detection step detects the intrusion of the vehicle. A vehicle detection method according to claim 8 or claim 9.

According to an eleventh aspect of the present invention, a program for realizing a vehicle detection method for detecting the vehicle in a target area in the image based on an image obtained by photographing a monitoring target range in which the vehicle travels on a road surface is recorded. In the recording medium, a pixel in a predetermined first luminance range in the target region of the image, a road surface region detection function of detecting as a road surface region, a predetermined second in the target region of the image Pixels within the luminance range, a dark area detection function to detect as a dark area, a road surface area detected by the road area detection function, and a temporal relationship between the dark area detected by the dark area detection function, or The spatial relationship is collated, and as a result of the collation, the dark region is determined to be the vehicle's shadow region only if the road surface region is connected temporally behind or continuously behind the dark region. And a recording medium of a vehicle detection method characterized by recording a program for realizing a lower shadow determination function for detecting the vehicle than this.

According to a twelfth aspect of the present invention, there is provided the recording medium according to the eleventh aspect, wherein the image is a spatio-temporal image in which certain areas are stacked with time.

According to a thirteenth aspect of the present invention, a program for realizing a vehicle detection method for detecting the vehicle in a target area in the image based on an image of a monitoring target range where the vehicle travels on a road surface is recorded. A change pattern extraction function for setting a vehicle intrusion detection area in the image and extracting a temporal change pattern of luminance of the intrusion detection area on the recording medium; and the change pattern extracted by the change pattern extraction function. Brightness level, dark brightness,
The road surface luminance, a luminance level classification function for classifying into at least three classes of other luminance, and the change pattern classified by the luminance level classification function, continue the dark luminance level for a predetermined duration or more, A recording medium for a vehicle detection method characterized by recording a program for realizing a vehicle intrusion detection function of detecting that a vehicle has entered when the road surface luminance level has changed.

According to a fourteenth aspect of the present invention, a program for realizing a vehicle detection method for detecting a vehicle in a target area in the image based on an image of a monitoring target range where the vehicle travels on a road surface is recorded. In the recording medium, the intrusion detection area of the vehicle is set in the image, a spatiotemporal image is generated by stacking the intrusion detection areas with time, and the spatiotemporal image generation function is generated by the spatiotemporal image generation function. A brightness level classification function for classifying the luminance level of the spatiotemporal image into at least three classes of dark luminance, road surface luminance, and other luminance; and a dark luminance level of the spatiotemporal image classified by the luminance level classification function. Detects that a vehicle has entered if the area lasts longer than a predetermined duration and then the road surface brightness level area connects in the direction immediately after the time. A recording medium of a vehicle detection method characterized by recording a program for realizing the vehicle intrusion detection that, the.

According to a fifteenth aspect of the present invention, the vehicle intrusion detecting function detects an intrusion of the vehicle based on an image at a time when the function detects the intrusion of the vehicle.
The recording medium of the vehicle detection method according to claim 13 or 14, further comprising a vehicle intrusion verification function for verifying the intrusion of the vehicle.

According to the invention having the above-described structure, since the false lower shadow area is a part of the vehicle body, an area other than the road surface area is connected next to the dark luminance area, but the true lower shadow area is dark. Only the true shadow region can be extracted using the fact that the road surface region connects after the luminance region.

[0030]

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

[First Embodiment] FIG. 1 is a block diagram showing a first embodiment of the present invention.

FIG. 2 is a flowchart showing the flow of processing of the vehicle detection device according to the first embodiment.

The vehicle detecting device includes image input means 11, binarizing means 12, road surface area extracting means 13, binarizing means 14,
It comprises a dark area extraction means 15, a lower shadow area determination means 16, and a result output means 17. These means can be realized by a computer such as a personal computer, and a program for realizing the means is a hard disk, an FD, a CD-ROM, or the like.
It is recorded on a recording medium such as a ROM and an MO.

The image input means 11 inputs an input image such as a camera or a video, or an image recorded in a storage device such as a hard disk of a computer, converts the input image into a digital image, and converts the image into a digital image. Is transmitted to the second binarizing means 14 (step 201).

The first binarizing means 12 extracts pixels within a predetermined luminance range and sets the pixels within the road luminance range to "1",
Pixels outside the road surface luminance range are binarized as “0” or the like.
The binarized data is transmitted to the road surface area extracting means 13 (Step 202).

A method of setting the predetermined luminance range will be described.

The digital image is generally processed as a 256-step gray scale image (256-step luminance image). When a color image is input, grayscale processing may be performed in advance. In brief, only the R component of the three components of RGB (red, green, blue) may be handled.

By the way, normally, the luminance of the road surface is within a certain width in the 256 levels. Generally, in a range that is neither bright nor dark, for example, a luminance of 40
It can be experimentally obtained in advance with a range such as between and 80. Generally, surveillance cameras for traffic flow monitoring have an automatic iris adjustment mechanism, and the brightness of the road surface changes over time due to changes in lighting, weather, and time, even at the same point, so a certain width It is important to determine whether the road surface luminance or not is the range as a range.

The brightness range can be set to the same brightness range for the entire screen. However, from a more general perspective, it is also effective to set a luminance range for each pixel position.

The binary data transmitted in this way is subjected to labeling processing by the road surface area extracting means 13. As a result, a road surface area in the image is extracted.
That is, it is cut out with the label information (step 203).

On the other hand, the second binarizing means 14 extracts pixels within a predetermined luminance range different from that of the binarizing means 12,
Pixels in the dark luminance range are “1”, pixels outside the range are “0”,
And binarize. The binarized data is transmitted to the road surface area extracting means 15 (Step 204).

Here, the luminance range of the dark area is, for example,
It is determined that the luminance is in the range of 0 to 40.

The upper limit of the brightness range of the dark area (4 in the above example)
0) and the lower limit of the luminance range of the road surface area (4 in the above example).
0) does not need to be the same value, but it is more erroneous to use the same value when detecting the shadow of the vehicle using the temporal change in luminance as described in the second embodiment. It is possible to reduce the number of different or indeterminate situations. Generally, they can be considered equivalent.

The binary data transmitted in this manner is subjected to labeling processing by the dark area extracting means 15. As a result, a dark area in the image is extracted. That is, it is cut out with the label information (step 20).
5).

In the conventional method, this dark area is extracted as it is as a shadow area, but in the present invention, a discrimination process based on connectivity is added.

The lower shadow area discriminating means 16 discriminates a lower shadow from the extraction results of the road area extracting means 13 and the dark area extracting means 15. If the road surface area is connected behind the dark area as a condition, the dark area is determined to be the shadow of the vehicle (step 206).

The connection conditions will be described in detail with reference to FIGS.

FIG. 3 is a schematic diagram showing an image of a surveillance camera in a tunnel. The installation direction of the camera in FIG. 3 is a direction in which the vehicle is monitored from behind, and the installation height of the camera is almost the same as the height of the tunnel. FIG. 3 shows an example in which a general vehicle is included in the input image.

As shown in FIG. 3, in a general vehicle, even if the color of the vehicle is gray or black, a shadow of the vehicle darker than the vehicle exists on the road surface. While the car body is relatively bright when illuminated by the lighting inside the tunnel, the under shadow is almost completely blocked by the car body, so the under shadow is darker than the body color.

In such a case, since the road surface luminance area is connected behind the lower shadow of the vehicle (the lower part of the lower shadow area in FIG. 3), the connection condition is satisfied. The dark area is determined to be a vehicle (step 206).

FIG. 4 shows an example in which a vehicle having a rear window is included in the input image. The rear window has a property of reflecting illumination light. Depending on the positional relationship between the lighting position and the vehicle, the light beam from the rear window may not reach the photographing camera, and may look very dark. Also, even when the inside of the vehicle body is dark, the rear window looks very dark. Even if it is not the rear window, depending on the design of the vehicle, black window frames and parts may look very dark.

In the case as shown in FIG. 4, the output of the dark area extracting means 15 is a rear window area and a true lower shadow area. According to the determination conditions of the present invention, the rear portion of the rear window is connected to the bright portion of the vehicle body and is not connected to the road surface brightness region. Therefore, the dark region of the rear window is determined not to be the shadow region. (Step 206).

On the other hand, since the true shadow of the vehicle is connected to the area of the road surface luminance, this dark area is determined to be the shadow area (step 206).

The result of the above determination is output from the result output means 17, and the process ends (step 207). And
For example, the position of the determined lower shadow area can be determined as the position of the vehicle body.

[Second Embodiment] In the first embodiment, an example in which an input image is processed as it is has been described. Instead, a vehicle detection method using spatiotemporal image processing will be described in the second embodiment.

When a vehicle enters from below the image as shown in FIGS. 3 and 4, the traffic flow monitoring system wants to detect the entry immediately when the vehicle enters. Therefore, the use of spatiotemporal image processing is effective.

FIG. 5 is a schematic diagram illustrating a method for generating a spatiotemporal image.

When processing a moving image as in traffic flow monitoring, the image is input as a sequence as time progresses as shown in FIG. Processing the entire screen of an image requires a large amount of processing, and real-time processing is difficult. In addition, in the case of the examples of FIGS. 3, 4, and 5A, the vehicle enters from below the screen.

Therefore, as shown in FIG. 5A, a cutout area is set in the lower part of the screen in the horizontal direction. The width of the cut-out area is equal to the width of the input image, but the vertical length is generally one pixel. As shown in FIG. 5B, when the elongated images of the cutout area are stacked with time, a new image is generated. This is a spatiotemporal image.

FIG. 6 shows an image sequence in which a vehicle actually intrudes is changed to a spatiotemporal image. The configuration of the vehicle detection device for the spatiotemporal image and the processing procedure are the same as those of the first embodiment.

FIG. 6 shows a spatio-temporal image separately extracted into a road surface area and a dark area in the same manner as in the first embodiment, and the result is represented by a schematic diagram. Since the region extraction result is classified into three, a road surface region, a dark region, and other regions, it may be considered that the spatiotemporal image is divided into three classes.

The method of determining the spatio-temporal image as shown in FIG. 6 based on the connection conditions in step 206 in FIG. 2 will be described in detail.

FIG. 6A shows an extraction result when a vehicle having a bright body passes. In the spatiotemporal image, the downward direction is the new time, and the upward direction is the old time. Since the vehicle body first passes through the cutout area in FIG. 5 and then the lower shadow passes, the result is as shown in FIG. 6A. In this case, since the road surface area (the area of the road surface brightness) is connected behind the extracted dark area (the area of the lower shadow), this dark area is determined as the lower shadow area.

FIG. 6B shows an extraction result when a vehicle having a dark body passes as in FIG. In this case, the vehicle body is also extracted as a dark area, but immediately after that, a dark area due to (true) under shadow is connected. The dark area of the vehicle body is determined to be "not a lower shadow area" because the road surface area is not connected to the rear. The dark region due to the (true) shadow is determined to be "a shadow region" because the road surface region is connected behind.

FIG. 6C shows an extraction result when a vehicle having a dark rear window passes as in FIG. In this case, a dark area due to a dark rear window and a dark area due to a (true) shadow are extracted. The dark area due to the dark rear window is not connected to the area of the bright vehicle body (the area that is not the road surface area) at the rear, and therefore does not satisfy the undershadow condition.
It is not determined as a shadow. On the other hand, the dark area due to the (true) shadow is determined to be "a shadow area" because the road area is connected to the rear.

In the second embodiment, a case where the present invention is applied to a spatiotemporal image has been described. The spatiotemporal image is suitable for quickly detecting the intrusion of the vehicle, but has a problem that the size of the vehicle in the spatiotemporal image changes depending on the speed of the vehicle. As can be seen from FIGS. 5 and 6, a vehicle with a high speed is projected short in the time direction, and a vehicle with a slow speed is projected long in the time direction. Therefore, it is difficult to detect the vehicle body based on the amount of the edge component or to detect the characteristic points of the vehicle body. What is certain is that the vehicle is detected from the information on the temporal change in luminance as shown in FIG. The temporal scale of the temporal change changes according to the speed of the vehicle, but the pattern of the change is constant. In this sense, the present invention using the pattern of time change as the determination condition is suitable for application to a spatiotemporal image.

Another advantage of applying the spatiotemporal image is that the lateral position and width of the invading vehicle can be detected as shown in FIG.

[Third Embodiment] In the second embodiment, a vehicle detection method using spatiotemporal image processing has been described. In the third embodiment, a method for simplifying spatiotemporal image processing and detecting a vehicle intrusion from a temporal change pattern of luminance will be described.

The connection condition will be reviewed as a temporal change in luminance. FIG. 7 is a graph showing a temporal change in luminance. Specifically, it may be considered that the change in luminance at the position A in FIG. 5A is graphed, or after generating a spatiotemporal space as shown in FIG. 6, a representative value at each time is extracted for each lane. It may be considered that the time change is graphed. In FIG. 6, the representative value is first divided at the position of the white line on the road surface, and the area having the largest width at the same time for each lane (in FIG. 6A, the largest width is obtained). In the area (a), a representative value (using an average value of the areas, etc.) of “other areas”, “dark area”, and “road surface area” may be used in chronological order.

FIG. 7A is a graph showing a change in luminance when a vehicle having a bright vehicle body passes as in FIG. The brightness range of the shadow and the brightness range of the road surface can be set as ranges having a certain width as shown in FIG. 7 (this width may be different for each pixel position). In the case of FIG. 7, the upper limit of the brightness range of the shadow is equal to the lower limit of the brightness range of the road surface.

Looking at the graph in the order of time, the luminance initially fluctuates slightly within the luminance range of the road surface. When the vehicle enters, the brightness changes to a brighter direction. Subsequently, when the lower shadow of the vehicle passes, the brightness changes suddenly and indicates a value in the brightness range of the shadow. After passing through the lower shadow, it returns to the luminance range of the road surface again.
According to the connection condition of the present invention, since the road surface luminance range is connected after the shadow luminance range (dark region luminance range), this lower shadow is determined to be a lower shadow.

FIG. 7B is a graph showing a change in luminance when a vehicle having a dark rear window passes, similarly to FIG. 6C.

Looking at the graph in order of time, the brightness initially fluctuates slightly within the brightness range of the road surface. When the vehicle enters, the brightness changes to a brighter direction. Subsequently, when the dark rear window passes, the luminance changes suddenly, and indicates the luminance in the luminance range of the shadow. Subsequently, the bright vehicle body passes again, and the brightness changes suddenly again to a bright brightness level. When the true lower shadow passes, the luminance again indicates a value within the luminance range of the shadow. After passing through the lower shadow, it returns to the luminance range of the road surface again.

According to the connection condition of the present invention, since the brightness connected to the rear of the rear window is the brightness of the bright part of the vehicle body, the area of this rear window is determined to be "not a shadow area", and the true shadow area is determined. Since the luminance range of the road surface is connected behind the lower shadow area, the lower shadow is determined to be “lower shadow area”.

The vehicle is detected by passing through the lower shadow area.

However, as described in the third embodiment, when a vehicle intrusion is detected from a temporal change pattern of one point in an image or a representative luminance at each time,
Unlike the case where the spatio-temporal image is used, the lateral position and the lateral width of the invading vehicle cannot be detected.

[Fourth Embodiment] As described above, FIG.
In the case of, the upper limit of the brightness range of the shadow is equal to the lower limit of the brightness range of the road surface. However, the determination method of the present invention is effective even when the upper limit of the shadow luminance range is not equal to the lower limit of the road surface luminance range. In the fourth embodiment, this case will be described.

FIG. 8 is a graph showing the change in luminance when the luminance range of the gray vehicle body is set between the luminance range of the road surface and the luminance range of the shadow (dark area). In FIG. 8, dots are clearly indicated on the graph data so that the sampling points on the time axis become clear.

For example, when a vehicle having a dark rear window and a gray car body passes as shown in FIG. 4, a change in luminance is as shown in FIG. The brightness of the gray body is lower than the brightness range of the road surface, but higher than the brightness range of the shadow. When passing through the rear window, the luminance falls within the gray luminance range but does not return to the road luminance range. After the true lower shadow area has passed, the value of the road surface luminance range is obtained at the next sample point. According to the determination conditions of the present invention, even in such a case, only the true lower shadow is determined as the lower shadow area.

[Fifth Embodiment] In the above-described embodiments, the intrusion of the vehicle from the spatial area arrangement of the input image or the temporal change pattern of the luminance including the spatiotemporal image processing is considered. Was detected. However, the spatial area arrangement (spatial pattern) and the temporal change pattern of the luminance have a characteristic that the advantages and the weak points can be mutually complemented.

In order to detect a vehicle without delay, it is advantageous to detect a vehicle using a spatiotemporal image or a luminance change pattern with emphasis on a temporal change. However, there is a possibility that excessive erroneous extraction may occur only with a temporal change. On the other hand, if the spatial arrangement of the image area (for example, the arrangement of the characteristic points of the vehicle or the pattern of the brightness distribution of the vehicle) is used, the result detected by the spatio-temporal image processing and the temporal change pattern of the luminance can be obtained. It becomes possible to verify.

FIG. 9 is a schematic diagram showing an example of detecting a vehicle intrusion using a spatiotemporal image and subsequently verifying the vehicle intrusion by comparing spatial patterns.

As described in the second embodiment, in the detection method based on the spatiotemporal image, the vehicle is detected immediately after the under shadow of the vehicle completely passes through the detection line. FIG. 9 shows the input image at that time.

The position and lateral width of the invading vehicle are detected by spatiotemporal image processing. Therefore, the rear surface pattern of the vehicle should have this position and the width, and fall within the range of the rectangle of the maximum height of the vehicle assumed from the width. If the vehicle moves to the left as time progresses as in the example of FIG. 9, the rectangular area may be set slightly to the left in anticipation of that.

In any case, the image for verifying the back pattern can be limited in terms of time and area.
Therefore, there is an advantage that the processing amount can be reduced. This is an advantage that cannot be realized by a device or a method for detecting the intrusion of a vehicle using only a spatial pattern, and can be obtained only by using information of both a temporal pattern and a spatial pattern.

There are many well-known methods for determining a vehicle using a spatial pattern. For example,
There is also a method of registering some simple template patterns as the back of the vehicle, obtaining the template pattern and a normalized correlation value, and increasing the similarity, thereby verifying that the image area is a vehicle. Although the template matching method is sensitive to the size conversion of the target and needs sufficient measures, the width of the vehicle is detected by spatiotemporal image processing. Even if the detection accuracy of the lateral width of the vehicle by the spatiotemporal image processing is not sufficient for the template matching, the processing can be simplified because the size of the vehicle can be limited as a range.

As a method for determining a vehicle by using a spatial pattern, the first embodiment of the present invention may be used. Since the image as the information to be determined is different between the spatiotemporal image and the image for verification, such an application is also possible.

[0088]

As described above, according to the present invention, the shadow of a vehicle can be accurately extracted, so that an image taken in a dark environment with a low installation height like a surveillance camera in a tunnel can be obtained. From the vehicle. Therefore, it is possible to provide a vehicle detection device that operates even in an installation environment or an installation angle of view to which the conventional method has not been applied. As a result, it can play an important role in realizing a monitoring system that automatically measures a traffic flow from a surveillance camera on a highway or detects an unexpected event automatically. This practical effect is truly significant.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a vehicle detection device according to an embodiment of the present invention.

FIG. 2 is a flowchart of a process of a vehicle detection device.

FIG. 3 is a schematic diagram of an input image.

FIG. 4 is a schematic diagram of an input image of a vehicle having a dark rear window.

FIG. 5 is a schematic diagram of a method for generating a spatiotemporal image.

FIG. 6 is a schematic diagram of an input image in the case of a spatiotemporal image.

FIG. 7 is a schematic diagram of a temporal change in luminance.

FIG. 8 is a schematic diagram of a temporal change in luminance when a luminance range of a gray vehicle body is set.

FIG. 9 is a schematic diagram of an example in which a detection result of spatio-temporal image processing is verified by matching a spatial pattern.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 Image input means 12 Binarization means 13 Road surface area extraction means 14 Binarization means 15 Dark area extraction means 16 Under shadow area discrimination means 17 Result output means

 ──────────────────────────────────────────────────続 き Continuing from the front page (72) Inventor Kazunori Onoguchi 8-6-26 Motoyama Minami-cho, Higashinada-ku, Kobe-shi, Hyogo F-term (reference) 2F065 AA01 BB05 BB15 CC11 FF01 FF04 JJ03 JJ19 JJ26 LL30 QQ04 QQ06 QQ31 QQ36 UU05 5B057 AA16 BA02 CA01 CA02 CA08 CA12 CB06 CB12 CC01 CE11 CE12 DA06 DB02 DB05 DB06 DB09 DC14 5H180 AA01 BB13 BB15 CC04 DD01

Claims (15)

[Claims] 1. A vehicle detection device for detecting a vehicle in a target area in an image based on an image of a monitoring target range in which the vehicle travels on a road surface, wherein the predetermined area in the target area in the image is determined. A road surface area detection unit that detects pixels in the first luminance range as a road surface area, and a dark area that detects pixels in a predetermined second luminance range in the target area of the image as a dark area. The detection unit, a road surface region detected by the road surface region detection unit, and a temporal relationship between the dark region detected by the dark region detection unit, or a spatial relationship, and as a result of the comparison, the dark region The temporal rear of, or, only when the road surface area is continuously connected to the spatial rear, the dark area is determined as the vehicle's shadow area, an under shadow determination unit that detects the vehicle from this, Characterized by consisting of Vehicle detection device. 2. The vehicle detection device according to claim 1, wherein the image is a spatiotemporal image in which certain areas are stacked with time progress. 3. A vehicle detection device for detecting the vehicle in a target area in an image based on an image of a monitoring target range in which the vehicle travels on a road surface, wherein an intrusion detection area of the vehicle is included in the image. And a change pattern extraction unit that extracts a temporal change pattern of the luminance of the intrusion detection area, and sets the luminance level of the change pattern extracted by the change pattern extraction unit to dark luminance, road luminance, A luminance level classifying unit for classifying into at least three classes of luminance, and the change pattern classified by the luminance level classifying unit continues a dark luminance level for a predetermined duration or more, and subsequently changes to a road surface luminance level. A vehicle detection device comprising: a vehicle intrusion detection unit that detects that a vehicle has entered the vehicle. 4. A vehicle detection device for detecting a vehicle in a target area in an image based on an image of a monitoring target range in which the vehicle travels on a road surface, wherein an intrusion detection area of the vehicle is included in the image. And a spatio-temporal image generation unit that generates a spatio-temporal image in which the intrusion detection areas are stacked with time progression. A luminance level classifying unit that classifies the luminance into at least three classes of other luminance, and a dark luminance level region of the spatiotemporal image classified by the luminance level classifying unit continues for a predetermined duration or more, and then A vehicle detection device comprising: a vehicle intrusion detection unit that detects that a vehicle has entered when a road surface luminance level region is connected in a direction immediately after time. 5. A vehicle intrusion verification unit that verifies the intrusion of the vehicle based on an image at a time when the vehicle intrusion detection unit detects the intrusion of the vehicle. The vehicle detection device according to any one of the preceding claims. 6. A vehicle detection method for detecting a vehicle in a target area in an image based on an image of a monitoring target range in which the vehicle travels on a road surface, the method comprising: A road surface area detecting step of detecting pixels in the first luminance range as a road surface area; and a dark area detecting pixels in a predetermined second luminance range in the target area of the image as a dark area. The detecting step, the road surface area detected by the road area detecting step, the temporal relationship between the dark area detected by the dark area detecting step, or the spatial relationship is compared, as a result of the comparison, the dark area Only when the road surface area is connected temporally behind or continuously behind the road, the dark area is determined to be the lower shadow area of the vehicle, and the lower shadow determination step for detecting the vehicle based on the dark area is determined. Vehicle detection method comprising the flop, to become more. 7. The vehicle detection method according to claim 6, wherein the image is a spatiotemporal image in which certain regions are stacked with time. 8. A vehicle detection method for detecting the vehicle in a target area in the image based on an image obtained by photographing a monitoring target range in which the vehicle travels on a road surface, wherein the vehicle intrusion detection area is included in the image. And a change pattern extraction step of extracting a temporal change pattern of the luminance of the intrusion detection area; and a luminance level of the change pattern extracted by the change pattern extraction step, a dark luminance, a road surface luminance, and the like. A luminance level classification step of classifying into at least three classes of luminance; and the change pattern classified by the luminance level classification step continues a dark luminance level for a predetermined duration or more, and subsequently changes to a road surface luminance level. A vehicle intrusion detecting step of detecting that a vehicle has entered in the case 9. A vehicle detection method for detecting the vehicle in a target area in an image based on an image of a monitoring target range in which the vehicle travels on a road surface, wherein the intrusion detection area of the vehicle is included in the image. And generating a spatio-temporal image in which the intrusion detection areas are stacked with time progress, and generating a spatio-temporal image, and calculating the luminance level of the spatio-temporal image generated by the spatio-temporal image generating step as dark luminance, road luminance. A luminance level classification step of classifying the luminance into at least three other classes; and a dark luminance level region of the spatiotemporal image classified by the luminance level classification step continues for a predetermined duration or more. A vehicle intrusion detection step of detecting that a vehicle has entered when the road surface luminance level region is connected in a direction immediately after the time. Vehicle detection method according to claim. 10. A vehicle intrusion verification step for verifying an intrusion of the vehicle based on an image at a time when the vehicle intrusion detection step detects the intrusion of the vehicle. The vehicle detection method as described in the above. 11. A recording medium storing a program for implementing a vehicle detection method for detecting the vehicle in a target area in the image based on an image obtained by photographing a monitoring target range in which the vehicle travels on a road surface. A pixel in a predetermined first luminance range in the target region of the image, a road surface region detection function of detecting as a road surface region, a pixel in a predetermined second luminance range in the target region of the image A, dark area detection function to detect as a dark area, a road surface area detected by the road area detection function,
The temporal relationship with the dark area detected by the dark area detection function, or the spatial relationship is collated, and as a result of the collation, the road area is continuously behind the dark area temporally or spatially behind. Only when the vehicle is connected, the dark area is determined to be a shadow area of the vehicle, and a shadow determination function of detecting the vehicle based on the dark area is recorded. Medium. 12. An image according to claim 1, wherein said image is a spatiotemporal image in which certain areas are stacked with time.
A recording medium for the vehicle detection method according to claim 1. 13. A recording medium storing a program for implementing a vehicle detection method for detecting a vehicle in a target area in an image based on an image of a monitoring target range in which the vehicle travels on a road surface. A change pattern extraction function of setting an intrusion detection area of the vehicle in an image and extracting a temporal change pattern of luminance of the intrusion detection area, and a luminance level of the change pattern extracted by the change pattern extraction function, A luminance level classification function for classifying into at least three classes of dark luminance, road surface luminance, and other luminance; and the change pattern classified by the luminance level classification function continues the dark luminance level for a predetermined duration or longer. And a vehicle intrusion detection function for detecting that a vehicle has entered when the road surface brightness level subsequently changes. Recording medium of vehicle detection method characterized by recording the beam. 14. A recording medium storing a program for implementing a vehicle detection method for detecting the vehicle in a target region in the image based on an image of a monitoring target range in which the vehicle travels on a road surface, the recording medium comprising: A spatio-temporal image generation function that generates a spatio-temporal image by setting the intrusion detection area of the vehicle in an image, stacking the intrusion detection area with time, and a spatio-temporal image generated by the spatio-temporal image generation function. A luminance level classification function for classifying the luminance level into at least three classes of dark luminance, road surface luminance, and other luminance; and a dark luminance level region of the spatiotemporal image classified by the luminance level classification function. Vehicle intrusion detection that detects that a vehicle has intruded if it lasts for more than the duration, and then the road surface brightness level area connects in the direction immediately after the time Recording medium of vehicle detection method characterized by recording a program for realizing the ability to. 15. A vehicle intrusion verification function for verifying the intrusion of the vehicle based on an image of the time at which the vehicle intrusion detection function detects the intrusion of the vehicle. A recording medium for the vehicle detection method according to any one of the preceding claims.