JP4381394B2 - Obstacle detection device and method - Google Patents

Description

  The present invention mainly uses a camera mounted on a vehicle to detect an obstacle existing on a road, such as a preceding vehicle, a parked vehicle, or a pedestrian, in order to realize safe driving assistance and automatic driving of the vehicle. The present invention relates to an obstacle detection apparatus and method thereof.

  As a method for detecting an obstacle present on the road, there are a method using an active sensor such as a laser range finder, an ultrasonic sonar, and a millimeter wave radar, and a method using a passive sensor such as a visible light TV camera or an infrared camera.

  Active sensors are used to measure the position of an object for various purposes, and its usefulness is widely known, but it detects objects that obstruct the running of the vehicle such as other vehicles on the road. For purposes, there are problems such as low detection resolution, insufficient measurement range, false detection of objects other than obstacles on the road, and erroneous detection of roadside objects that do not become obstacles because the driving lane cannot be detected. is there. Therefore, an advanced obstacle detection technique based on image analysis using a passive sensor such as a TV camera is desired so as to solve these problems.

  It can be said that the method of detecting an obstacle by analyzing an image from a TV camera or the like mounted on the vehicle generally detects the obstacle or recognizes the traveling lane from the luminance pattern information of the image. Even from the image of one camera, it is possible to detect a traveling lane by cutting out a gray portion with little texture.

  However, since there are actually many obstacles with brightness and patterns similar to roads, it is difficult to realize a highly practical one with few false detections.

  On the other hand, there is a method of detecting an obstacle or a traveling lane using a plurality of cameras, which is generally referred to as “stereo vision”.

In stereo vision, 3D information of the detection target area can be obtained, so it is expected to realize more accurate obstacle detection and lane detection, but in the real world shown between multiple camera images In general, there is a problem that a correspondence search for finding the same point cannot be solved uniquely. In this regard, the methods such as Patent Document 1 and Patent Document 2 do not require a correspondence search and are very useful for realizing lane detection and the like.
JP 2001-76128 A JP 2000-293893 A

  However, when parallax is seen in the roadside guardrails, textures such as white lines and signs on the road surface due to slopes and pitching of the host vehicle, these guardrails and road surface textures do not become obstacles to driving, There was a problem that it was detected as an obstacle. In rainy weather, raindrops adhering to the windshield and camera lens surface are also erroneously detected as obstacles.

  Therefore, the present invention has been made in view of such circumstances, by comparing features obtained in a plurality of camera images, or by calculating an index value indicating whether the image features are derived from an obstacle. The object of the present invention is to reduce the erroneous detection as described above and greatly improve the performance of the obstacle detection device.

The present invention, in the obstacle detection device for detecting whether an obstacle exists within the range taken by the imaging apparatus, an image from the imaging device and an image input means for entering, with respect to the image, a plurality of elongated Element area setting means for setting the rectangular element areas to be arranged in order in the horizontal direction, feature calculation means for calculating the direction of the luminance gradient of each element area and its frequency distribution , and each element for each element area An index calculation means for calculating an index value by obtaining an element index from the direction of the luminance gradient in the area and its frequency distribution, and summing up the element indexes of each element area; and if the index value is equal to or greater than a threshold, the failure It is an obstacle detection apparatus characterized by having the result determination means which determines with an object.
Further, the present invention includes a plurality of imaging devices arranged in parallel to each other, images a common range by the plurality of imaging devices, and detects whether there is an obstacle in the common range. In the obstacle detection device, a feature amount of an area having the same size composed of an image input unit for inputting an image from the imaging device and a plurality of pixels in a predetermined image with respect to the plurality of input images is calculated. When the feature calculation unit, the feature comparison unit that calculates the difference between the calculated feature amounts of the plurality of images, the difference between the calculated feature amount and a threshold value are compared, and the difference between the feature amounts is smaller than the threshold value, Feature amount combining means for combining the feature amounts of the images, and element region setting means for setting a plurality of vertically long element regions in order in the horizontal direction to the feature amounts obtained by combining the feature amounts of the images. And before An index calculation unit that calculates an index value indicating the presence or absence of the obstacle from the distribution of the feature amount for each element region, and a result determination unit that determines the presence or absence of an obstacle from the calculated index value This is an obstacle detection device.

  According to the present invention, an object that does not obstruct traveling such as a roadside guardrail, a white line on a road surface, or a texture such as a sign is erroneously detected as an obstacle, or adheres to a windshield or a camera lens surface in rainy weather. It is possible to reduce the error of the obstacle detection device such that the raindrop to be erroneously detected as an obstacle and greatly improve the performance.

(First embodiment)
Hereinafter, the obstacle detection apparatus 10 according to the first embodiment of the present invention will be described with reference to FIGS.

  FIG. 1 is a schematic configuration diagram of an obstacle detection apparatus 10 of the present embodiment.

  The obstacle detection device 10 includes an image input unit 11, a feature calculation unit 12, a feature comparison unit 13, and a result determination unit 14.

  Each function of these units 11 to 14 can be realized by a program stored in a computer such as a personal computer.

(1) Image input unit 11
The image input unit 11 is supplied with an image from an arbitrary imaging device 101. The imaging device 101 has imaging devices such as a plurality of TV cameras, and is attached to two places on the left side and the right side of the front part of the host vehicle, for example, as shown in FIG.

  The image input unit 11 sequentially A / D-converts video signals continuously output from the imaging device or the like, and stores them in a memory in the image input unit 11 as a digital image or a time series image. The requested arbitrary time and arbitrary region image are output to the next feature calculation unit 12.

(2) Feature calculation unit 12
The feature calculation unit 12 calculates a predetermined feature amount for an arbitrary time / arbitrary region of the image or image sequence sent from the image input unit 11.

  The feature amount indicates arbitrary information that can be obtained in each pixel or arbitrary region.

  For example, from simple things such as pixel brightness values and color information, any object detection or region such as spatial or temporal differentiation or integration results, arbitrary filter superposition results, statistics such as average or variance, etc. It is possible to use an arbitrary scalar quantity or vector quantity that can be calculated from the image, such as the result of the division processing and the feature quantity and motion quantity of the region obtained from the result.

  Here, a case where the luminance gradient at the pixel position is used as the feature amount will be described.

  The luminance gradient indicates the degree of luminance change in the vicinity of the pixel, and takes a large value in the boundary region of the object shown in the image and the structure in the object (also simply referred to as an edge).

  As a method for obtaining the luminance gradient, for example, there is a method of superimposing a filter such as a Sobel operator on an image. The method will be briefly described with reference to FIG.

When the luminance value at the position (x, y) in the image is represented by I (x, y) and each filter element value Fx (i, j) is set as shown in the left diagram of FIG. Value obtained by superimposing the filter on the image

Increases when there is a change in luminance in the horizontal direction in the vicinity of (x, y), so that a luminance gradient in the horizontal direction can be obtained from Dx.

Similarly, when the filter Fy (i, j) is set as shown in the right diagram of FIG.

To obtain a luminance gradient in the vertical direction.

From these,
The intensity gradient intensity is (Dx ² + Dy ² ) ^1/2 ,
The luminance gradient direction is tan ⁻¹ Dy / Dx,
Etc.

  There are various other filters for obtaining the luminance gradient, such as Laplacian, Robinson, and Canny, and any of them may be used.

  Further, it may be obtained by a statistical method such as Hough transform.

  There are many known brightness gradient calculation methods, for example, detailed in Reference Document 1 (supervised by Takagi and Shimoda: Image Analysis Handbook, University of Tokyo Press, ISRN 4-13-061107-0).

(3) Feature comparison unit 13
The feature comparison unit 13 extracts only the features derived from the obstacle from the image by comparing the feature amounts of the images from the plurality of imaging devices obtained by the feature calculation unit 12.

  4 and 5 are schematic diagrams of processed images. 4 and 5 are examples of images obtained from the cameras installed as shown in FIG. It is assumed that the camera is mounted facing forward in parallel with the vehicle body traveling direction.

  FIG. 4 is an example when there is an obstacle ahead (the preceding vehicle in this example), the left figure is an example of an image from a camera (left camera) installed on the left side of the host vehicle, and the right figure is the right side. It is an example of an image from the camera (right camera).

  FIG. 5 shows an example when there is no obstacle. Similarly, the left figure is an image from the left camera, and the right figure is an image from the right camera.

  Now, in the left and right images, it is assumed that candidate areas to be subjected to obstacle detection processing are set as indicated by bold rectangles R1 and R2 in each figure, and whether there are obstacles in these candidate areas. The purpose is to determine whether or not.

  As shown in FIG. 4, when an obstacle is included in the candidate area, if the candidate area is correctly set in the parallax according to the distance between the camera and the obstacle, the candidate area is derived from the obstacle in the left and right candidate areas. A similar luminance pattern (in this example, the back of the preceding vehicle) is seen. Here, “parallax” refers to “displacement” of the position of an object image in an image caused by a difference in position between cameras when an object is captured by a plurality of cameras. In the example of this figure, since the optical axes of the left and right cameras are installed parallel to the traveling direction of the host vehicle, the position of the preceding vehicle image in the right figure is shifted to the right side from the position in the left figure. It corresponds to that.

  Furthermore, in this example, if it can be assumed that the distance between the camera and the obstacle is sufficiently larger than the distance difference between the points inside the obstacle (the unevenness on the back of the preceding vehicle), the parallax inside the obstacle is ignored. It can be considered that the luminance pattern in the image derived from the obstacle between the images of the left and right cameras is simply translated with a substantially constant parallax.

  On the other hand, when there is no obstacle as shown in FIG. 5, there is a difference in the luminance pattern seen in the left and right candidate areas.

  This is because the spread in the depth direction of an object that is not an obstacle in the candidate area (in this figure, a white line on the road surface, a structure attached to a road such as a guard rail) cannot be ignored with respect to the distance between the camera and the object. This is because the parallax within the object changes greatly according to the distance.

  A feature comparison method for selectively outputting an image feature of an obstacle having a plane perpendicular to the traveling direction of the vehicle using the difference in appearance due to the parallax is, for example, In this way, it can be realized.

  Now, assuming that the candidate area R1 set in the left camera image and the candidate area R2 in the right camera image are set to the same size, the pixel position with reference to R1 and the pixel with reference to R2 The position can be expressed in common, and this is (x, y).

If the feature amount at the pixel position (x, y) calculated by the feature calculation unit 12 is F1 and F2 in the left camera image and the right camera image, respectively, the difference in the feature amount at this position (x, y) is Amount shown

If only those having a small difference in feature amount are output, only the features derived from the obstacle can be output.

  The parallax changes greatly according to the distance, such as a white line or a guardrail, which does not become an obstacle to traveling in normal driving because it has a large depth in the traveling direction of the host vehicle. Therefore, a large difference appears in the “angle” of the projected image of the object between the left and right camera images. On the other hand, since the obstacle image of the preceding vehicle or the like only moves substantially parallel between the left and right camera images, there is almost no difference in image angle.

Therefore, in order to distinguish the two, it can be said that one of the effective methods is to obtain the direction of the luminance gradient of the object image in the images of the left and right cameras and use the difference. Assuming that the luminance gradient is obtained by the feature calculation unit 12 as in the above example, the luminance gradient direction at (x, y) of the left camera image is d1 (x, y), and the right camera image is Let d2 (x, y) be

It is an example of the simplest feature amount difference.

Also, the Euclidean distance between feature vectors f1 (x, y) and f2 (x, y) whose elements are luminance gradient strength and direction.

Can be used, a geometrical difference between feature quantities can be used, or can be calculated from arbitrary feature quantities F1 (x, y) and F2 (x, y) calculated by the feature calculation unit 12. Any amount can be used as the feature amount difference D.

  The feature comparison unit 13 calculates a feature amount difference D for an arbitrary pixel, and outputs this value itself or an amount uniquely determined for D for an arbitrary pixel (x, y).

(4) Result determination unit 14
The result determination unit 14 uses the image feature derived from the obstacle obtained by the feature comparison unit 13 to determine whether an obstacle exists in the image or the image area to be determined.

The feature comparison unit 13 outputs a large pixel value I (x, y) when the feature amount difference D (x, y) is small in the pixel (x, y) in the candidate region. If the pixels for which no feature is obtained are ignored, the sum of I in the candidate region R

It is possible to determine the presence or absence of an obstacle by obtaining.

  When there is an obstacle such as a vehicle in the front, S is large because there are many pixels with a small difference D of the feature amount. Conversely, when there is no obstacle such as a vehicle and only the guardrail or the road surface can be seen in the image Has a small feature amount difference D, and S is small.

  Therefore, by determining the threshold th and comparing this with the sum S, it is possible to easily determine whether there is an obstacle.

  As a method for performing the determination, any value that can be calculated using the difference D between the feature amounts obtained by the feature comparison unit 13 can be used.

  As a result of determination by the result determination unit 14, if it is determined that there is an obstacle, the alarm device 102 may give an alarm to the driver, or the vehicle body control device 103 may apply braking or steering. May be performed. Further, the result may be output or transferred to an arbitrary device such as the communication device 104.

  Note that the threshold th does not have to be constant for all pixels in R and all times, and may be given a different value for each pixel or each time by a predetermined method, or any value obtained from an image It is also possible to change based on the quantity (scalar quantity or vector quantity).

(5) Obstacle detection in case of rain Here, the obstacle detection in case of rain is considered.

  If raindrops adhere to the windshield of the vehicle or the lens surface of the camera on which it is mounted, conventional obstacle detection may cause these raindrop images to be mistakenly detected as obstacles, or the obstacle detection results to be incorrect. There was a high possibility that an error would occur.

  On the other hand, in the case of comparing feature quantities between a plurality of camera images as in the first embodiment, the probability of chance that similar raindrops simultaneously attach to a location corresponding to a specific parallax in all camera images (coincidence) ) Is small, it can be seen that there is a sufficiently high probability that the difference in the feature amount of the image portion to which raindrops have adhered will be larger between the camera images, and erroneous detection due to raindrop images can be effectively reduced.

(6) Modification 1
In the first embodiment, the candidate area is rectangular. However, the candidate area can have any predetermined shape.

  In addition, the description has been made assuming that the candidate areas R1 and R2 in the left and right camera images are determined in advance. However, an obstacle detection method such as Japanese Patent Application No. 2001-154569 is used as a preprocessing, and this result is used as a candidate area. It is also possible to set any candidate lane by using a predetermined method based on the result of using any driving lane detection method.

(7) Modification 2
Furthermore, even if the candidate area is not determined in advance, if the operation of the first embodiment is performed at each scan position by changing the position in the left and right camera images, the same obstacle as a result Detection processing can be performed.

  In this way, when an obstacle is detected by analyzing an image obtained by an imaging device such as a TV camera, an object that does not actually interfere with traveling, such as a road surface texture or a guardrail reflected in the image. Without erroneous detection, only obstacles such as preceding vehicles can be detected correctly, and malfunctions such as issuing false alarms or performing unnecessary vehicle control can be greatly reduced.

(Second Embodiment)
An obstacle detection device 20 according to a second embodiment of the present invention will be described with reference to FIGS.

  FIG. 6 is a schematic configuration diagram of the obstacle detection device 20 of the present embodiment.

  The second embodiment includes an image input unit 21, a feature calculation unit 22, an index calculation unit 23, and a result determination unit 24.

  As in the first embodiment, the image input unit 21 is supplied with an image from an arbitrary imaging device 101 or the like. In the second embodiment, since the feature amounts of images obtained from a plurality of imaging devices are not compared, the imaging device 201 is not necessarily a plurality, and may be a single one.

  Further, the image input unit 21 and the feature calculation unit 22 in the present embodiment can be the same as those in the first embodiment.

(1) Index calculation unit 23
The index calculation unit 23 uses the image feature amount obtained by the feature calculation unit 22 to calculate an index indicating whether the image feature is derived from an obstacle.

  7 and 8 are schematic diagrams of processed images.

  As in the first embodiment, it is assumed that the candidate region R is preset in the image, and the luminance gradient is used as the image feature. FIG. 7 is an example of the candidate region R when an obstacle (in this example, a vehicle ahead) is included, and FIG. 8 is an example when there is no obstacle.

  The index calculation unit 23 obtains an index P used for result determination from element indices Pn obtained in a plurality of element areas Cn (n = 1... N) set in advance in the candidate area R, and obtains the next result. It outputs to the determination part 24.

  A method for calculating the index P and a method for using the index P will be described with reference to FIG.

(1-1) Setting of Element Area Cn As shown in FIG. 7, the element area Cn is set as five vertically long rectangles in the candidate area R of the image.

(1-2) Calculation of element index Pn The element index Pn is obtained from the frequency distribution (histogram) in the luminance gradient direction in each element region Cn. C1, C3, and C5, which are examples of this histogram, are shown in FIG.

  The luminance gradient direction θ plots the appearance frequency f (θ) of the luminance gradient in the element region Cn where θ is −90 ° to 90 °, with the horizontal being 0 ° and the clockwise direction being positive.

  As a result, the histogram shows the direction distribution of luminance gradients (edges) often found in each element region Cn.

  For example, in C1 and C5, both ends of the vehicle are shown and vertical edges are noticeable, so the value of f (90) or f (−90) is large.

  On the other hand, since C2 to C4 include an image of the center of the vehicle and a horizontal edge is noticeable, the value of f (0) increases.

  That is, the element index Pn is determined in advance as Pn = f (90) or Pn = f (−90) when n = 1, 5, and Pn = f (0) when n = 2, 3, or 4. In this case, the value of the element index Pn increases in all element regions only when there is an obstacle.

  Here, as a method for determining the element index Pn, there are the following methods.

  The first method is a method in which the value of f (θ) at a certain θ determined for each Pn as described above is used as it is.

  The second method is a method of taking a statistic such as a maximum value or an average value of f (θ) within a predetermined range of θ.

  For example, the range of θ is −90 ° = <θ−70 ° and 70 ° = <θ = <90 ° when n = 1, 5, and when n = 2, 3, 4 -30 ° = <θ = <30 °.

(1-3) Calculation of index P If the histogram is normalized by the total cumulative frequency, this can be regarded as a probability distribution in the edge direction.

Therefore, if the index P is obtained as a joint likelihood that there is an obstacle,

It can be.

(1-4) Method of Using Index P When the index P is obtained in this way, each element index value Pn is almost 0 in any case of n in the case of FIG. 8 where there is no obstacle. , P is also almost zero.

  On the other hand, since P when there is an obstacle is larger than 0, this P is an index indicating whether the image feature is derived from the obstacle, that is, an index indicating whether there is an obstacle. Can be used for determination.

  By the same method, it is possible to calculate an index P ′ for detecting an object that does not cause an obstacle such as a guardrail or a white line on the road.

  For example, it is assumed that an area including only C1 and C2 in FIG. 8 is set as the candidate area R ′.

If the luminance gradient direction distribution of C1 and C2 is a distribution having a peak near α ° (α ° is about 50 ° as seen from the figure), Pn = f (α) and n = About 1 and 2

Can be used as an index indicating that there is a guardrail or white line on the left side of the road.

  In exactly the same manner, an index indicating the presence of a guardrail or a white line on the right side of the road can be calculated. For any object, an index indicating the presence of the object can be calculated based on the characteristic distribution unique to the object.

(2) Result determination unit 24
The result determination unit 24 uses the index P indicating that there is an obstacle obtained by the index calculation unit 23 to determine whether there is an obstacle in the image or the image area to be determined.

  The presence or absence of an obstacle may be determined by setting a certain threshold th and comparing it with an index P. As described above, the index P that there is an obstacle and an object other than the obstacle The presence / absence of an obstacle may be determined by comparing the indicators P ′, P ″,.

  As a result of the determination by the result determination unit 24, the result may be transmitted to an arbitrary device such as the reporting device 202, the vehicle body control device 203, or the communication device 204.

  In this way, it is possible to correctly detect only obstacles such as preceding vehicles, and to generate false alarms, distinguishing them from objects that do not actually interfere with traveling such as road surface patterns and guardrails shown in the image. Malfunctions such as unnecessary vehicle control can be greatly reduced.

(3) Modification 1
As a method for performing the determination, any value that can be calculated using any index P obtained by the index calculation unit 23 can be used.

(4) Modification 2
In the second embodiment, the element region C is set to five rectangular shapes having the same size, but can be set to an arbitrary number and an arbitrary shape so that a part thereof overlaps. May be set.

(5) Modification 3
Further, although only the luminance gradient distribution has been described as the image feature, an arbitrary feature amount or feature vector distribution obtained from the feature calculation unit can be used.

(6) Modification 4
As for the element index Pn, the method of using the frequency value f (θ) at a predetermined angle θ according to n has been described, but as a method for obtaining Pn, for example, the mode value in a certain angle range, the average An arbitrary statistic such as a value, median, or weighted sum, or a value obtained by an arbitrary calculation formula of f (θ) can be set.

(7) Modification 5
A description will be given of how to obtain a difference in feature amount when there are three or more imaging devices.

  For example, if the variance value of the feature amounts of w images is obtained, the variation of the w feature amounts can be indicated, and the variation of the feature amounts can be used as the difference of the feature amounts.

(Third embodiment)
An obstacle detection device 30 according to a third embodiment of the present invention will be described.

  FIG. 9 is a schematic configuration diagram of the third embodiment of the present invention.

  The third embodiment includes an image input unit 31, a feature calculation unit 32, a feature comparison unit 33, an index calculation unit 34, and a result determination unit 35.

  The image input unit 31, the feature calculation unit 32, and the feature comparison unit 33 are the same as those in the first embodiment, and the index calculation unit 34 and the result determination unit 35 are the same as those in the second embodiment. .

  In order to perform the feature comparison unit 33 and the index calculation unit 34 continuously, the following function is added to the feature comparison unit 33.

  That is, at arbitrary positions (x, y) in the candidate regions R1 and R2 in the images of the left and right cameras, the feature amounts in the candidate regions R1 and R2 are f1 (x, y) and f2 (x, y), respectively. If it is determined that the feature amount difference D is smaller than the threshold th, the feature amounts f1 (x, y) and f2 (x, y) are combined to generate the feature amount f (x, y). ) Is output, otherwise 0 is output.

As a method of synthesizing the feature quantity f, for example, it is assumed that f, f1, and f2 are all M-dimensional vector quantities (in the case where the brightness gradient is used as the feature, the brightness gradient strength and direction are elements of the vector. ), Each element fm of f (m = 1, 2, 3,..., M)

fm (x, y) = min {f1m (x, y), f2m (x, y)}

As described above, the minimum value may be obtained for each element.

  As another synthesis method, any value calculated from f1 (x, y) and f2 (x, y) can be used.

  In this way, by continuously performing the feature comparison unit 33 and the index calculation unit 34, it is possible to correctly detect only obstacles more reliably and to significantly reduce malfunctions such as issuing false alarms and performing unnecessary vehicle control. Can be made.

It is a schematic block diagram of the obstacle detection apparatus of the 1st Embodiment of this invention. It is a schematic diagram which shows the installation state of an imaging device. It is a schematic diagram which shows the method of calculating | requiring a brightness | luminance gradient in a feature calculation part. It is a schematic diagram explaining the process in the characteristic comparison part when there exists an obstruction. It is a schematic diagram explaining the process in the characteristic comparison part when there is no obstacle. It is a schematic block diagram of the obstruction detection apparatus of the 2nd Embodiment of this invention. It is a schematic diagram explaining the process in the parameter | index calculation part when there exists an obstruction. It is a schematic diagram explaining the process in the parameter | index calculation part when there is no obstruction. It is a schematic block diagram of the obstruction detection apparatus of the 3rd Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Obstacle detection apparatus 11 Image input part 12 Feature calculation part 13 Feature comparison part 14 Result determination part 20 Obstacle detection apparatus 21 Image input part 22 Feature calculation part 23 Index calculation part 24 Result determination part 30 Obstacle detection apparatus 31 Image input Unit 32 feature calculation unit 33 feature comparison unit 34 index calculation unit 35 result determination unit

Claims (7)

In the obstacle detection device that detects whether there is an obstacle within the range imaged by the imaging device,
An image input means image is input from the imaging device,
With respect to the image, an element area setting means for setting a plurality of vertically long element areas to be arranged in order in the horizontal direction;
Feature calculating means for calculating the direction of the luminance gradient of each element region and its frequency distribution ;
Index calculation means for obtaining an element index from the direction of the luminance gradient in each element area and its frequency distribution for each element area, and calculating the index value by summing the element indexes of each element area ;
A result determination means for determining that the obstacle is present if the index value is equal to or greater than a threshold ;
An obstacle detection device comprising: In the obstacle detection device that includes a plurality of imaging devices and is arranged in parallel to each other, images a common range by the plurality of imaging devices, and detects whether or not an obstacle exists in the common range ,
Image input means for inputting images from the imaging device;
Feature calculating means for calculating a feature amount of a region of the same size composed of a plurality of pixels in a predetermined image with respect to the plurality of input images ;
A feature comparison means for obtaining a difference in the feature values for the plurality of calculated images;
Comparing the obtained feature quantity difference with a threshold value, and, when the feature quantity difference is smaller than the threshold value, synthesizing the feature quantity of each image;
Element region setting means for setting a plurality of vertically long rectangular element regions in order in the horizontal direction to the feature amount obtained by combining the feature amounts of the images ;
Index calculation means for calculating an index value indicating the presence or absence of the obstacle from the distribution of the feature amount for each element region;
A result determination means for determining the presence or absence of an obstacle from the calculated index value;
An obstacle detection device comprising: The feature amount in the feature calculation means is a luminance gradient direction or luminance gradient intensity of the region,
The index calculation means obtains an element index from the combined feature amount and frequency distribution in each element region, calculates the index value by summing the element indexes of each element region,
The obstacle detection device according to claim 2 , wherein the result determination unit determines that the obstacle is present if the index value is equal to or greater than a threshold value. In the obstacle detection method for detecting whether or not an obstacle exists within the range imaged by the imaging device,
An image input step of image is input from the imaging device,
Regarding the image, an element area setting step for setting a plurality of vertically long element areas to be arranged in order in the horizontal direction;
A feature calculating step for calculating the direction of the luminance gradient of each element region and its frequency distribution ;
An index calculation step for obtaining an element index from the direction of the luminance gradient in each element area and its frequency distribution for each element area, and calculating the index value by summing the element indices of each element area ;
A result determination step for determining that the obstacle exists if the index value is equal to or greater than a threshold ;
An obstacle detection method characterized by comprising: In the obstacle detection method in which a plurality of imaging devices are arranged in parallel to each other, a common range is photographed by the plurality of imaging devices, and whether or not an obstacle exists in the common range is detected.
An image input step in which images are respectively input from the imaging device;
A feature calculating step of calculating a feature amount of a region of the same size composed of a plurality of pixels in a predetermined image with respect to the plurality of input images ;
A feature comparison step for obtaining a difference between feature amounts of the plurality of calculated images;
A feature amount combining step of comparing the obtained feature amount difference with a threshold value and combining the feature amounts of the images when the feature amount difference is smaller than the threshold value;
An element region setting step for setting a plurality of vertically long element regions in order in the horizontal direction to the feature amount obtained by combining the feature amounts of the images ;
An index calculation step for calculating an index value indicating the presence or absence of the obstacle from the distribution of the feature value for each element region;
A result determination step for determining the presence or absence of an obstacle from the calculated index value;
An obstacle detection method characterized by comprising: The obstacle detection program for detecting whether an obstacle exists within the range taken by the imaging device,
On the computer,
An image input function of the image is input from the imaging device,
With respect to the image, an element area setting function for setting a plurality of vertically long element areas to be arranged in order in the horizontal direction;
A feature calculation function for calculating the direction of the luminance gradient of each element region and its frequency distribution ;
An index calculation function for obtaining an element index from the direction of the luminance gradient in each element area and its frequency distribution for each element area, and calculating the index value by summing the element indexes of each element area ;
A result determination function for determining that the obstacle is present if the index value is equal to or greater than a threshold ;
Obstacle detection program for realizing. The imaging device is arranged in parallel with each other a plurality, a common range photographed by the plurality of imaging devices, in the obstacle detection program for detecting whether an obstacle within the common range is present ,
On the computer,
An image input function for inputting images from the imaging device;
A feature calculation function for calculating a feature amount of a region of the same size composed of a plurality of pixels in a predetermined image with respect to the plurality of input images ;
A feature comparison function for obtaining a difference between feature amounts of the plurality of calculated images;
A feature amount combining function for comparing the obtained feature amount difference with a threshold value and combining the feature amounts of the images when the feature amount difference is smaller than the threshold value;
An element region setting function for setting a plurality of vertically long rectangular element regions in order in the horizontal direction to the feature amount obtained by combining the feature amounts of the images ;
An index calculation function for calculating an index value indicating the presence or absence of the obstacle from the distribution of the feature amount for each element region;
A result determination function for determining the presence or absence of an obstacle from the calculated index value;
Obstacle detection program for realizing.

Images