JP2015088029A - Radial distortion correction device, road environment recognition device, radial distortion correction method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To easily and automatically correct radial distortion of a camera without using an especial calibration device such as a checker board.SOLUTION: A radial distortion correction device includes: image input means 11 which inputs a captured image photographed by an imaging device being an object of radial distortion correction; temporary distortion correction means 12 which supposes two or more distortion correction parameters for the input captured image to generate corresponding temporary distortion corrected images; local gradient direction histogram calculation means 13 which calculates a local gradient direction histogram for each of the generated temporary distortion corrected images; bias calculation means 14 which calculates a degree of bias in local gradient directions in each of the temporary distortion corrected images on the basis of the local gradient direction histogram; distortion corrected image selection means 15 which selects one distortion corrected image from the temporary distortion corrected images on the basis of degrees of bias in local gradient directions in respective temporary distortion corrected images.

Description

  The present invention relates to a radial distortion correction apparatus, a radial distortion correction method, a radial distortion correction program, and a road environment recognition apparatus using the same.

  As a method for correcting the radial distortion of the camera, for example, there is a method described in Non-Patent Document 1. The method described in Non-Patent Document 1 is a distortion in which a checkerboard or the like is placed in front of a camera to capture an image and an image correction program is used to satisfy the geometrical requirements in which the coordinates of the lattice points are determined. A correction coefficient is estimated, and image conversion according to the estimated distortion correction coefficient is applied.

  As a method not using a special calibration marker such as a checkerboard, there is a method described in Non-Patent Document 2. In the method described in Non-Patent Document 2, an object including a linear shape such as a pillar or a shelf is imaged, and a curve corresponding to the object including such a linear shape is captured in a captured image using a contour tracking method called “snakes”. And the distortion correction coefficient is estimated from the position information of the curves.

Z. Zhang, "A flexible new thechnique for camera calibration", IEEE Trans. On Pattern Analysis and Machine Intelligence, Vol.22 (11), 2000, p.1330-1334. S.B.Kang, "Radial Distortion Snakes", IAPR Workshop on Machine Vision Applications, 2000.

  However, if the method described in Non-Patent Document 1 or Non-Patent Document 2 is used to automatically correct the radiation distortion of an in-vehicle camera or the like, there are the following problems.

  That is, when using the method described in Non-Patent Document 1, a checker board is installed each time on the assembly line of a product (for example, a vehicle) on which the camera is mounted, or the product is removed from the production line and a wide camera It has to be set by moving to a calibration field and calculating distortion correction parameters. In either case, complicated work is required. Also, since a special calibration device such as a checkerboard is required, there is a problem that calibration is not easy even if the optimal value of the distortion correction parameter may change after the product has been handed over, such as when the vehicle is running. there were. In the case of a vehicle, if the calibration work is neglected, the road environment recognition accuracy is adversely affected. Therefore, a method that allows easy calibration is desired.

  Further, when the method described in Non-Patent Document 2 is used, there is a problem that it is necessary for a human to give an initial contour, and calibration cannot be automatically performed.

  Therefore, in view of the above points, the present invention provides a radial distortion correction apparatus and a radial distortion correction method that can easily and automatically correct the radial distortion of a camera without using a special calibration device such as a checkerboard. An object of the present invention is to provide a radial distortion correction program.

  It is another object of the present invention to provide a road environment recognition device that can maintain the recognition accuracy of the road environment even when the optimum value of the distortion correction parameter changes during vehicle travel.

  The radial distortion correction apparatus according to the present invention includes an image input unit that inputs a captured image captured by an imaging device that is a correction target device for radial distortion, and two or more distortions with respect to the captured image input by the image input unit. Assuming correction parameters, a temporary distortion correction unit that generates a temporary distortion correction image corresponding to each assumed distortion correction parameter, and each of the temporary distortion correction images generated by the temporary distortion correction unit A local gradient direction histogram calculating means for calculating a local gradient direction histogram representing a distribution of local gradient directions, which is a gradient direction in a local region in the image, and a local gradient direction histogram calculated by the local gradient direction histogram calculating means. Bias calculation means for calculating the degree of bias in the local gradient direction in each of the temporary distortion correction images, and bias calculation Distortion correction image selection means for selecting one distortion correction image from the temporary distortion correction images based on the degree of bias in the local gradient direction in each of the temporary distortion correction images calculated by the stages. It is characterized by.

  The road environment recognition apparatus according to the present invention includes an image input unit that inputs a captured image captured by an imaging device installed in a vehicle, and two or more distortion correction parameters for the captured image input by the image input unit. Assuming that a temporary distortion correction unit that generates a temporary distortion correction image corresponding to each assumed distortion correction parameter and a temporary distortion correction image generated by the temporary distortion correction unit Based on the local gradient direction histogram calculation means for calculating the local gradient direction histogram representing the distribution of the local gradient direction, which is the gradient direction in the local region, and the local gradient direction histogram calculated by the local gradient direction histogram calculation means, A bias calculation means for calculating the degree of bias in the local gradient direction in each of the provisional distortion correction images, and a bias calculation means; Distortion correction image selection means for selecting one distortion correction image from the provisional distortion correction images based on the degree of bias in the local gradient direction in each of the calculated temporary distortion correction images, and distortion correction image selection means A distortion correction parameter determining unit that determines a distortion correction parameter to be applied to a captured image input after the current time based on a selection result of the distortion correction image by the distortion correction parameter determined by the distortion correction parameter determination unit Is applied to captured images input after the current time to generate a distortion-corrected image, and a distortion-corrected image generated by the distortion-corrected image generating means is subjected to image recognition processing. And image recognition means for recognizing the road environment around the vehicle.

  The radial distortion correction method according to the present invention inputs a captured image captured by an imaging device that is a target device for correcting radial distortion, assumes two or more distortion correction parameters for the input captured image, and assumes A temporary distortion correction image corresponding to each distortion correction parameter generated is generated, and a distribution of local gradient directions, which are gradient directions in local regions in the image, is represented for each of the generated temporary distortion correction images. A local gradient direction histogram is calculated, and based on the calculated local gradient direction histogram, a degree of bias in the local gradient direction in each of the temporary distortion correction images is calculated, and a local gradient in each of the calculated temporary distortion correction images is calculated. One distortion correction image is selected from provisional distortion correction images based on the degree of bias in the gradient direction.

  The radial distortion correction program according to the present invention includes an image input process for inputting a captured image captured by an imaging device that is a correction target device for radial distortion to a computer, and two or more distortion correction parameters for the input captured image. And a temporary distortion correction process for generating a temporary distortion correction image corresponding to each of the assumed distortion correction parameters, and for each of the generated temporary distortion correction images, in a local region in the image A local gradient direction histogram calculation process for calculating a local gradient direction histogram representing a distribution of local gradient directions, which is a gradient direction. Based on the calculated local gradient direction histogram, the local gradient direction bias in each of the temporary distortion correction images is calculated. Based on the bias calculation processing for calculating the degree, and the degree of bias in the local gradient direction in each of the calculated temporary distortion correction images. Te, characterized in that to perform the distortion correction image selection processing for selecting one of the distortion correction image from the temporary distortion correction image.

  According to the present invention, the radial distortion of the camera can be corrected easily and automatically without using a special calibration device such as a checkerboard.

  Further, according to the present invention, it is possible to maintain the recognition accuracy of the road environment even when the optimum value of the distortion correction parameter changes during vehicle travel.

It is a block diagram which shows the structural example of the radial distortion correction apparatus of 1st Embodiment. It is a figure which shows an example of a captured image. It is a figure which shows the graph as an example of a local gradient direction histogram, and the value of entropy. Is a graph showing an example of the relationship between the distortion correction coefficient k ₁ and entropy value. It is a flowchart which shows the operation example of the radial distortion correction apparatus of 1st Embodiment. It is a figure which shows an example of the hardware constitutions of the radial distortion correction apparatus of 1st Embodiment. It is explanatory drawing for demonstrating the distance used for a weight. It is a block diagram which shows the structural example of the radial distortion correction apparatus of 2nd Embodiment. It is a flowchart which shows the operation example of the radial distortion correction apparatus of 2nd Embodiment. It is a block diagram which shows the structural example of the radial distortion correction apparatus of 3rd Embodiment. It is a figure which shows the example of the relationship between the value of a distortion correction parameter, and the value of entropy when entropy does not have a clear minimum value. It is a block diagram which shows the structural example of the road environment recognition apparatus of 4th Embodiment. It is a flowchart which shows the operation example of the road environment recognition apparatus of 4th Embodiment. It is a block diagram which shows the other structural example of the road environment recognition apparatus of 4th Embodiment. It is a flowchart which shows the other operation example of the road environment recognition apparatus of 4th Embodiment. It is a block diagram which shows the outline | summary of this invention.

Embodiment 1. FIG.
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration example of a radial distortion correction apparatus according to the first embodiment. The radial distortion correction apparatus shown in FIG. 1 corrects the radial distortion of an in-vehicle camera installed in a vehicle as an example. The radial distortion correction apparatus includes an imaging unit 101, a temporary distortion correction unit 102, a local gradient direction histogram calculation unit 103, an entropy calculation unit 104, and a distortion correction image selection unit 105.

  The imaging means 101 includes an imaging device such as an in-vehicle camera, which is a device to be corrected for radial distortion, and an interface unit that inputs an image captured by the imaging device. The obtained image (hereinafter referred to as a captured image) is input. For example, when the imaging unit 101 is installed in a vehicle, it may input an image based on a video signal obtained by photographing a surrounding scenery or the like that changes every moment such as a road or a building on its side. The image may be a color image or a monochrome image. Further, it may be a moving image or a still image.

  FIG. 2 is an explanatory diagram illustrating an example of a captured image. As shown in FIG. 2, in general, in an imaging device such as a camera that connects images using a lens, a phenomenon called “radiation distortion” in which a linear shape in the real world is captured in a barrel shape is captured. Occur. This occurs because a straight line is distorted and projected due to the distortion of the lens, and the distortion becomes more prominent as the image becomes more peripheral. Note that the example of the captured image shown in FIG. 2 is converted into a binarized image in order to emphasize the outline of the object shown in the image.

  The temporary distortion correction unit 102 performs distortion correction processing based on the temporary distortion correction parameter on the input captured image, and generates a temporary distortion correction image. The temporary distortion correction unit 102 generates at least two temporary distortion correction parameters.

If the pixel position of the captured image input from the imaging means 101 is (x, y) and the pixel position of the generated temporary distortion correction image is (x ′, y ′), the correspondence between the two is the distortion correction parameter. Using (k ₁ , k ₂ , k ₃ , k ₄ ,..., k _D ), the following expression is used.

  In fact, it is widely known that there is no problem in accuracy even if approximation is performed with D = 2. That is, there is no problem even if the above formula (1) is changed to the following formula (2).

The temporary distortion correction means 102 assumes various assumptions of distortion correction parameters (k ₁ ,..., K _D ) used in the distortion correction models shown in the above-described equations (1) and (2). Various temporary distortion correction images are generated by applying each distortion correction parameter to the distortion correction model. Note that the image before distortion correction and the image after correction may each be a color image or a monochrome image.

  The local gradient direction histogram calculation unit 103 calculates a local gradient direction histogram for each temporary distortion correction image generated by the temporary distortion correction unit 102. The local gradient direction histogram is a histogram whose measurement value is a gradient direction (hereinafter, referred to as a local gradient direction) that is a direction of change in luminance in a local partial region around an arbitrary pixel in an image. Specifically, the local gradient direction histogram only needs to express the distribution of the local gradient direction in the target image. In the present embodiment, as the local gradient direction histogram, the frequency (appearance number and frequency (probability)) for each class when the local gradient direction around each pixel obtained from the target image is divided into a predetermined class (bin). calculate. An example is introduced below. For example, the local gradient direction histogram calculation means 103 uses the local gradient direction θ (x, y) obtained for each pixel and the gradient magnitude m (x, y) as follows. A gradient direction histogram may be calculated. The gradient magnitude m (x, y) is used as a weight to be given to each local gradient direction θ when the frequency for each class is obtained.

  When the temporary distortion correction image is a color image, the local gradient direction histogram calculation means 103 performs the following processing after converting it into a luminance image. That is, for each pixel in the luminance image I, the local gradient direction θ (x, y) and the gradient magnitude m (x, y) are calculated by the following equation (3).

Here, the function F _x (x, y) is a function that returns a differential value of a pixel value (in this example, a luminance value) in the x-axis direction (horizontal direction) at the pixel position (x, y). The function F _y (x, y) is a function that returns a differential value of the pixel value in the y-axis direction (vertical direction) at the pixel position (x, y). In the case of a digitized image, the differential operation with respect to the change in the pixel value can be replaced with a difference operation. Therefore, the function F _x (x, y) and the function F _y (x, y) may be functions that perform a difference operation as shown in the following expressions (4) and (5). In addition, when performing a difference calculation, it is preferable to perform noise reduction and noise removal using a Prewitt filter, a Sobel filter, or the like.

For example, the local gradient direction histogram calculation means 103 uses the Prewitt operator,
F _x (x, y) = I (x + 1, y-1) + I (x + 1, y) + I (x + 1, y + 1) -I (x-1, y-1)- I (x-1, y) -I (x-1, y + 1)
F _y (x, y) = I (x-1, y + 1) + I (x, y + 1) + I (x + 1, y + 1) -I (x-1, y-1)- I (x, y-1) -I (x + 1, y-1)
... (4)
Or with less computation,
F _x (x, y) = I (x + 1, y) -I (x-1, y)
F _y (x, y) = I (x, y + 1) -I (x, y-1)
... (5)
It is good. Here, I (x, y) represents a luminance value at the pixel position (x, y).

  Next, the local gradient direction histogram calculation means 103 weights and averages the local gradient direction θ (x, y) obtained for each pixel for each predetermined direction section as shown in the following equation (6). Then, a local gradient direction histogram is calculated.

  Here, N represents the number of divisions in the local gradient direction, that is, the number of classes (bins) in the histogram. In addition, i represents the bin number, that is, the identifier of the direction segment in the local gradient direction. Therefore, h (i) represents the bin histogram value (frequency (probability) in this example) indicated by i. Δ (i, j) represents the Kronecker delta symbol. I represents the entire luminance image. M (x, y) is a weight. [] Represents a Gaussian symbol.

  The entropy calculation unit 104 indicates the degree of local gradient direction bias in each of the temporary distortion correction images generated by the temporary distortion correction unit 102 based on the local gradient direction histogram calculated by the local gradient direction histogram calculation unit 103. Calculate entropy as an indicator. For example, the entropy calculating unit 104 may calculate the entropy E of the local gradient direction histogram using the following equation (7) for each temporary distortion correction image.

  The value of entropy E approaches 0 as the histogram value h (i) obtained for each bin in the local gradient direction histogram of the target image is distributed toward a specific bin (ie, a specific direction segment). The more uniformly distributed, the larger the value.

  As an example, FIG. 3 shows a graph and an entropy value as an example of a local gradient direction histogram respectively obtained from an image with radial distortion and an image without radial distortion for the same scene. The horizontal axis of the graph shown in FIG. 3 shows bin, which is a direction segment in the local gradient direction divided every 5 degrees, and the vertical axis shows the histogram value, that is, the weighted frequency (probability) of the local gradient direction in the image. Yes. Of the two bar graphs described for each bin, the left side (black bar) shows the histogram value of the local gradient direction histogram obtained from the image without radial distortion, and the right side (white bar) shows the image with radial distortion. The histogram value of the local gradient direction histogram calculated | required from is shown.

  From the graph shown in FIG. 3, an image without radial distortion has a particularly high histogram value corresponding to the local gradient directions of 0 degrees, 90 degrees, and 180 degrees compared to an image with radial distortion, It can be seen that the bias is large. FIG. 3 also shows the entropy value of each local gradient direction histogram. As shown in FIG. 3, the entropy value of the local gradient direction histogram of the image without radial distortion is 3.803, and the entropy value of the local gradient direction histogram of the image with radial distortion is 3.966. It can be seen that the entropy value is small.

FIG. 4 is a graph showing changes in entropy values when the degree of distortion correction is variously changed for the same image. The graph of FIG. 4 shows how entropy values change when the distortion correction coefficient k ₁ , which is a parameter included in the distortion correction parameters, is variously changed. Comparing the degree of distortion of the distortion-corrected image and the entropy value when the distortion correction coefficient k ₁ is changed, it can be seen that the entropy value 401 for an image without radial distortion takes a minimum value. Similar graphs can be obtained even when other distortion correction coefficients other than the distortion correction parameters are changed. Thus, it is understood that the radial distortion is corrected well in the image that gives the minimum value of the entropy value.

  The distortion correction image selection unit 105 selects one distortion correction image based on the entropy value calculated for each of the temporary distortion correction images generated by the temporary distortion correction unit 102. Specifically, the distortion correction image selecting unit 105 selects a temporary distortion correction image that gives a minimum entropy value from among the generated temporary distortion correction images as a distortion correction image with the most radial distortion corrected. . As a result, it is possible to obtain an estimation result that the temporary distortion correction parameter used to generate the selected image is the distortion correction parameter that most corrects the radial distortion. The radial distortion correction apparatus can also correct radial distortion for a captured image input thereafter using the distortion correction parameter estimated in this way.

  Next, the operation of the radial distortion correction apparatus of this embodiment will be described. FIG. 5 is a flowchart showing an operation example of the radial distortion correction apparatus of this embodiment. As shown in FIG. 5, first, the imaging unit 101 captures a scene or the like reflected in the installed state, and captures a captured image (step S101).

  Next, the temporary distortion correction unit 102 sets various temporary distortion correction parameters, performs temporary distortion correction on the captured image input using each of the set temporary distortion correction parameters, and outputs the corresponding temporary distortion correction parameters. The corrected image is generated (step S102).

  Next, the local gradient direction histogram calculation unit 103 calculates a local gradient direction histogram for each of the temporary distortion correction images generated by the temporary distortion correction unit 102 (step S103).

  Next, the entropy calculation unit 104 calculates the entropy of each local gradient direction histogram generated by the local gradient direction histogram calculation unit 103 (step S104).

  Finally, the distortion-corrected image selection unit 105 selects a temporary distortion-corrected image that gives a minimum entropy among the entropies calculated by the entropy calculation unit 104 as an image in which the radial distortion is correctly corrected (step S105). .

  As described above, according to the present embodiment, the imaging means that is the imaging device to be corrected without taking the trouble of installing the checkerboard on the assembly line each time or removing the vehicle from the production line and moving it. A good distortion correction parameter can be automatically estimated from the captured image taken by the image 101, and an appropriate distortion correction image can be obtained.

  This is based on the following principle. The entropy of the local gradient direction histogram takes a small value when the appearance frequency of the entire image in the local gradient direction defined for each pixel is biased. Further, a large number of straight lines existing in the real world form a straight line on a correctly corrected distortion image, and conversely form a curve on a distorted image. Since the local direction of a straight line or curve is orthogonal to the local gradient direction, the image that gives the local entropy minimum value of the local gradient direction histogram is an image in which the straight line shape in the real world is correctly converted to a straight line, That is, it can be said that the image has been correctly corrected for distortion.

  Further, according to the present embodiment, the distortion correction parameter can be satisfactorily optimized even with a camera in which the optimum value of the distortion correction parameter changes on the vehicle or the like. This is because the distortion correction parameter is not calibrated only once on the assembly line, but can be dynamically estimated and optimized while the vehicle is running. Therefore, if the radial distortion correction apparatus of this embodiment is installed in a vehicle or the like, it is possible to recognize the road environment satisfactorily.

  FIG. 6 is a diagram illustrating an example of a hardware configuration of the radial distortion correction apparatus according to the present embodiment. The radial distortion correction apparatus according to the present embodiment may be realized by an embedded system having a hardware configuration as shown in FIG. 6, for example. The embedded system shown in FIG. 6 includes an ECU (Electronic Control Unit) 601 and a camera 602. The ECU 601 controls the entire embedded system, and includes, for example, a CPU 603, a RAM 604, a ROM 605, a signal processing circuit 606, a power supply circuit 607, and the like. The ROM 605 of the ECU 601 is equipped with modules corresponding to the respective processing means of the radiation distortion correction apparatus described above, and the CPU 603 reads out and executes these modules, whereby the embedded system operates as a radiation distortion correction apparatus. In this example, the imaging unit 101 is realized by a camera 602 and a signal processing circuit 606. The temporary distortion correction unit 102, the local gradient direction histogram calculation unit 103, the entropy calculation unit 104, and the distortion correction image selection unit 105 are realized by a CPU 603 that operates according to a program. It is also possible to configure a microcomputer by implementing the functions executed by the ECU 601 as hardware. Furthermore, some functions may be realized by hardware, and similar functions may be realized by cooperative operation of the hardware and the software program.

  In the above embodiment, an example in which the local gradient direction histogram calculation unit 103 calculates the local gradient direction histogram using the luminance image is shown, but the image used for calculating the local gradient direction histogram is not limited to the luminance image. . For example, an image obtained by extracting only a green component value, a red component value, or a blue component value of a color image may be used. In such a case, the local gradient direction histogram is a local gradient direction θ (x, x, which represents the direction of change in pixel value (green component value, red component value, or blue component value) in a local region around any pixel in the image. It may be a histogram of y). Further, for example, an image having a pixel value that is a sum of a red component value, a green component value, and a blue component value may be used, or a near infrared image may be used. That is, the image used for the calculation of the local gradient direction histogram may be an image including brightness information in the pixel value.

In the above embodiment, D = 2, that is, an example in which the temporary distortion correction unit 102 corrects radial distortion using distortion correction parameters (k ₁ , k ₂ ) including two distortion correction coefficients is mainly shown. However, even when a higher-order distortion correction parameter is used, the amount of calculation increases, but an image in which the camera's radial distortion is suitably corrected can be obtained in exactly the same procedure.

  Further, in the above embodiment, an example is shown in which the gradient magnitude m (x, y) is used as a weight when the local gradient direction histogram calculation unit 103 forms a histogram of the local gradient direction obtained for each pixel. However, the weight is not limited to this. For example, g (m (x, y)) converted using a function g that preliminarily determines the magnitude m (x, y) of the gradient may be used as the weight. g may be, for example, a step function or a function that clips the upper limit of m (x, y). Also, for example, a weight obtained by multiplying the gradient magnitude m (x, y) by a value corresponding to the distance d (x, y) between the straight line orthogonal to the local gradient direction θ (x, y) and the image center is weighted. It is good.

  FIG. 7 is an explanatory diagram showing an example of the distance d (x, y) used for the weight. 7A is an explanatory diagram showing an example of the captured image, FIG. 7B is an outline-enhanced image of the captured image shown in FIG. 7A, and FIG. 7C is FIG. It is explanatory drawing which cuts out and shows the pin and line which were attached | subjected for description to the image shown to b).

  Now, as shown in FIG. 7B, the local gradient direction θ (x, y) in the pixel of interest 902 in the image 901 is a direction orthogonal to the white line on the road surface as in the direction 903. In this case, a distance 906 (see FIG. 7C) between the straight line 904 orthogonal to the direction 903 and the image center 905 corresponds to the distance d (x, y).

  When the distance d (x, y) is relatively small (for example, less than the overall average), the change in the corresponding bin value is small no matter how the distortion correction parameter is changed. In other words, such a local gradient direction of a pixel having a relatively small distance d (x, y) does not contribute much to increase or decrease of the entropy value. Therefore, the weight for the local gradient direction in such a pixel (that is, a pixel having an edge passing in the vicinity of the center of the image) is made relatively small, and an important pixel around the image (contribution to increase / decrease in entropy) By relatively increasing the weight of the local gradient obtained from a large pixel), the distortion correction parameter can be obtained more accurately.

  For example, the local gradient direction histogram calculation means 103 uses the following equation (8) instead of the above-described equation (6) to weight the average of the local gradient direction θ (x, y) of each pixel for each predetermined direction segment. Then, a local gradient direction histogram may be calculated. In Equation (8), a value obtained by multiplying the gradient magnitude m (x, y) by the distance d (x, y) is used as a weight for the measured value (local gradient direction for each pixel). Instead of using the distance d (x, y) as it is, a value obtained by multiplying the magnitude m (x, y) of the gradient by a value corresponding to the distance d (x, y) may be used as the weight.

  Further, in order to omit the calculation of the distance d (x, y), for each small area in the image, as a weight value derived from the distance corresponding to each predetermined direction section, for each predetermined direction section in the small area A weight table that holds values corresponding to the distance d (x, y) or the distance d (x, y) may be prepared. When calculating the local gradient direction histogram, the local gradient direction histogram calculation means 103 may obtain a weight value derived from a distance corresponding to each measurement value with reference to this weight table. In this way, the weight can be determined only by the table lookup process without calculating the square root.

  In the above-described embodiment, an example in which entropy is used as an index indicating the degree of bias in the local gradient direction in the image (hereinafter referred to as bias index) is shown, but the bias index is not limited to entropy. For example, other statistics (standard deviation, variance, sum of squares, etc.) obtained from the local gradient direction histogram may be used. When standard deviation, variance, or sum of squares is used as the bias index, standard deviation, variance, or sum of squares is used instead of entropy calculation means as a bias calculation means for calculating the degree of bias in the local gradient direction in each temporary distortion correction image. It is sufficient to provide means for calculating Further, the distortion corrected image selecting means 105 has the radial distortion corrected most for the temporary distortion corrected image that gives the maximum value among the statistics (standard deviation, variance, or sum of squares) obtained for each temporary distortion corrected image. Select an image. The values of variance, standard deviation, and sum of squares increase as the bias increases, so the maximum value can be obtained when the real-world straight line shape is converted into a straight line, that is, an image with correct distortion correction. Because it can be said.

  It should be noted that the bias calculation means can be used instead of entropy or standard deviation as long as it is an index indicating the distribution in the local gradient direction in the image and the frequency bias as a scalar value as the bias index. .

  In the above embodiment, an example is shown in which the local gradient direction histogram is calculated based on the Prewitt operator and the difference between the left and right and upper and lower pixel values, but the method of calculating the local gradient direction histogram is not limited to this. Instead, any method may be used as long as it obtains a gradient direction distribution in a local partial region around an arbitrary pixel in the target image.

Embodiment 2. FIG.
Next, a second embodiment of the present invention will be described. FIG. 8 is a block diagram illustrating a configuration example of the radial distortion correction apparatus according to the present embodiment. The irradiation distortion correction apparatus shown in FIG. 8 includes an imaging unit 201, a temporary distortion correction unit 202, a local gradient direction histogram calculation unit 203, an entropy calculation unit 204, a parameter sequential determination unit 205, and a distortion correction image generation unit. 206.

  The imaging unit 201, local gradient direction histogram calculation unit 203, and entropy calculation unit 204 are the same as the imaging unit 101, local gradient direction histogram calculation unit 103, and entropy calculation unit 104 in the radial distortion apparatus of the first embodiment shown in FIG. It has the same function as the function.

The temporary distortion correction unit 202 of the present embodiment assumes a plurality of different values for one of the unknown parameters (distortion correction coefficients) among the distortion correction parameters used in the distortion correction model. A distortion corrected image is generated. In the present embodiment, the temporary distortion correction means 202 is called at least for the number of unknown parameters. For example, each time the temporary distortion correction unit 202 is called, the temporary distortion correction unit 202 selects one of the parameters corresponding to the term having a smaller order of r, sets a plurality of values for the selected distortion correction coefficient, and sets the corresponding temporary A distortion corrected image may be generated. In this example, the distortion correction coefficient _{k 1, k 2, k 3} , k 4, are selected one by one in the order of ... k _D.

At this time, for example, when the distortion correction coefficient k ₁ is selected, the distortion correction coefficients k ₂ , k ₃ , k ₄ ,..., K _D that are unknown parameters corresponding to higher-order terms are fixed values, For example, 0 may be set. Further, after the value of the currently selected distortion correction coefficient k ₁ is determined by the parameter sequential determination unit 205 described later, the parameter corresponding to the low-order term is selected at the stage where the distortion correction coefficient k ₂ is selected by the next call. The distortion correction coefficient k ₁ is a value determined by the parameter sequential determination means 205, and the distortion correction coefficients k ₃ , k ₄ ,..., K _D that are unknown parameters corresponding to higher-order terms are fixed. Set a value, for example 0. On top of that, variously setting the current value of the distortion correction coefficient k ₂ is an unknown parameter of the selected, to generate a distortion correction image corresponding formal.

  The parameter sequential determination unit 205 selects a temporary distortion correction image that gives a minimum entropy based on the entropy for each temporary distortion correction image calculated by the entropy calculation unit 204, and is used to generate the temporary distortion correction image. A value set for the temporary distortion correction parameter that has been set is determined as an estimation result for one unknown parameter (distortion correction coefficient) that is currently selected. The parameter sequential determination unit 205 calls the temporary distortion correction unit 202 again when there are still unknown parameters, and generates various temporary distortion correction images corresponding to the next unknown parameters. In this way, the value of the unknown parameter is sequentially determined, and when the value of the last unknown parameter is determined, the distortion correction parameter composed of these values is changed to the distortion correction parameter for correcting the most radial distortion. It is assumed that Since the temporary distortion correction parameters selected in the final stage, that is, the final unknown parameter determination stage, have all the values determined in the previous stages, the parameter sequential determination means 205 The temporary distortion correction parameter selected in the last stage may be used as it is as the estimation result of the distortion correction parameter for correcting the most radial distortion.

  The distortion correction image generation unit 206 corrects the radial distortion of the input captured image using the distortion correction parameter for correcting the most radial distortion determined by the parameter sequential determination unit 205, and the radial distortion is detected. A correctly corrected distortion corrected image is generated.

  Next, the operation of this embodiment will be described. FIG. 9 is a flowchart showing an operation example of the radial distortion correction apparatus of the present embodiment. As shown in FIG. 9, first, the imaging unit 201 captures a captured image by capturing a scene or the like reflected in the installed state (step S <b> 201).

  Next, the temporary distortion correction unit 202 selects one of the unknown parameters among the distortion correction parameters, sets various values for the selected unknown parameter, and generates a corresponding temporary corrected image. (Step S202). In step S <b> 202, the temporary distortion correction unit 202 selects, in order, parameters corresponding to lower-order terms among unknown parameters.

  Next, the local gradient direction histogram calculation unit 203 calculates a local gradient direction histogram for each of the temporary distortion correction images generated in step S202 (step S203).

  Next, the entropy calculation unit 204 calculates entropy from the local gradient direction histogram of each temporary distortion correction image calculated in step S203 (step S204).

  Next, the parameter sequential determination unit 205 selects a temporary distortion correction image that gives a minimum entropy among the entropies calculated in step S204, and uses the temporary distortion correction image used to generate the temporary distortion correction image. The value set for the correction parameter is determined as an estimation result for one of the unknown parameters selected in step S202 (step S205).

  Next, when there are still unknown parameters (Yes in step S206), the process returns to step S202. On the other hand, if there are no unknown parameters (No in step S206), the process proceeds to step S207.

  In step S207, the distortion correction image generation unit 206 performs a conversion process on the input captured image using the distortion correction parameters including all the parameter values determined by the parameter sequential determination unit 205, so that the distortion is correct. A corrected distortion corrected image is generated (step S207). Note that since the distortion corrected image has already been selected for the captured image used when determining the distortion correction parameter, it is not necessary to generate a distortion corrected image again. The distortion correction image generation unit 206 may generate a distortion correction image for the captured image input after the distortion correction parameter is determined.

As described above, according to the present embodiment, the distortion correction parameter can be optimized at a higher speed. For example, when considering a case in which M values are set for each parameter of the distortion correction parameters (k ₁ ,..., K _D ) to obtain a combination that gives the minimum entropy, the configuration of the first embodiment in it it is necessary to search for the minimum value by setting the distortion correction parameters of the temporary street M ^D, according to the configuration of the present embodiment, requires only search for M × D Street. In general, since the lower-order terms have a greater influence on the radiation distortion, a good distortion correction parameter can be obtained using all the finally determined parameters by the method of sequentially determining the parameters corresponding to the lower-order terms. Can be obtained.

Embodiment 3. FIG.
Next, a third embodiment of the present invention will be described. FIG. 10 is a block diagram illustrating a configuration example of the radial distortion correction apparatus according to the present embodiment. As shown in FIG. 10, the radial distortion correction apparatus of this embodiment includes a distortion correction parameter determination means 305 instead of the parameter sequential determination means 205 in the radial distortion correction apparatus of the second embodiment shown in FIG. Is different.

  The distortion correction parameter determination unit 305 further includes a distortion correction parameter storage unit 3051 and a time series determination unit 3052 in addition to the parameter sequential determination unit 205 described above.

  The distortion correction parameter storage unit 3051 generates a distortion correction parameter value estimated for the captured image input from the imaging unit 201 every moment, more specifically, a temporary distortion correction image selected as the distortion correction image. The value of the distortion correction parameter used in is stored. The distortion correction parameter storage unit 3051 is finally determined by, for example, the value of the distortion correction parameter determined as the estimation result by the method described in the first or second embodiment, or the method of the present embodiment described later. The value of the distortion correction parameter may be stored together with the time information.

  The time series determination unit 3052 calculates the median of the distortion correction parameters from the distortion correction parameter group estimated for the captured images at the most recent predetermined number of times stored in the distortion correction parameter storage unit 3051, A distortion correction parameter composed of the calculated values is used as a final estimation result. For example, the time-series determination unit 3052 may regard the distortion correction parameter as a vector or matrix having D elements, and calculate the median of the values of such vectors or matrices, or for each distortion correction coefficient. The median value may be calculated. Note that the latest time may include the current time.

  By selecting the median of parameters obtained in time series, the distortion correction parameter can be obtained more stably.

  When time series processing is not performed, the estimated distortion correction parameter greatly depends on the properties of the input instantaneous video. For example, when a captured image that is an estimation target is unsuitable for estimating a distortion correction parameter, it is difficult to obtain a stable parameter value unless time series processing is performed. The case where the distortion correction parameter is not suitable is a case where an object having a linear structure is not included in the image. Therefore, a stable distortion correction parameter can be obtained by reducing the influence of the instantaneous video property.

  As a mechanism for excluding the distortion correction parameter estimated for the image unsuitable for the estimation of the distortion correction parameter, the distortion correction parameter storage unit 3051 takes a local minimum value with clear entropy as shown in FIG. If not, the distortion correction parameter estimated at that time may not be stored. FIG. 11A to FIG. 11C are graphs showing examples of the relationship between the distortion correction coefficient value having the distortion correction parameter and the entropy value when the entropy does not have a clear minimum value. The example shown in FIG. 11A is an example in which entropy increases monotonously with respect to a change in distortion correction coefficient value. Further, the example shown in FIG. 11B is an example in which the entropy decreases monotonously with respect to the change in the distortion correction coefficient value. Further, the example shown in FIG. 11C is an example in which the minimum value of entropy does not appear clearly with respect to the change in the distortion correction coefficient value. In this way, by excluding estimation results when the entropy does not take a minimum value or when the minimum value does not appear clearly in the generated temporary distortion correction image group, the distortion correction parameter can be more stably set. It is possible to estimate.

  Note that the maximum value of the number of distortion correction parameter groups stored in the distortion correction parameter storage unit 3051 is determined in advance. The time series determination unit 3052 finally selects, for example, the distortion correction parameter based on the median value obtained when the number of distortion correction parameter groups stored in the distortion correction parameter storage unit 3051 reaches the maximum value, The correction parameter estimation process may be terminated, or the distortion correction parameter storage unit 3051 may be configured as a FIFO structure, and the median value may be obtained every moment based on the moving median method to continuously update the distortion correction parameter. Good.

  In the above embodiment, an example is shown in which the distortion correction parameter storage unit 3051 and the time series determination unit 3052 are added to the configuration of the second embodiment to perform time series processing. Similar additions can be made to the configuration of one embodiment. In such a case, the distortion correction image selection unit 105 is replaced with the distortion correction parameter determination unit 305, and the distortion correction parameter determination unit 305 includes the distortion correction image selection unit 105, the distortion correction parameter storage unit 3051, and the time series determination unit. 3052 may be included.

Embodiment 4 FIG.
Next, a fourth embodiment of the present invention will be described. FIG. 12 is a block diagram illustrating a configuration example of the surrounding environment recognition apparatus of the present embodiment. The surrounding environment recognition apparatus shown in FIG. 12 further includes image recognition means 401 and display means 402 in addition to the configuration of the radial distortion correction apparatus of the third embodiment shown in FIG.

  The image recognition unit 401 recognizes a predetermined object from the image output from the distortion correction image generation unit 206 using an image recognition technique. For example, when the surrounding environment recognition device according to the present embodiment is a road environment recognition device that is mounted on a vehicle and used to recognize the road environment, the image recognition unit 401 uses the specified image as a pedestrian. Or a predetermined object such as an obstacle, a vehicle, or a road sign, and safe driving support such as obstacle detection, vehicle detection, road sign recognition, or the like may be performed. For example, the image recognition unit 401 may recognize a predetermined object from the image, specify the type of the object, or calculate the distance between the vehicle and the object.

  In the present embodiment, the image recognition unit 401 performs image recognition processing on the distortion corrected image output from the distortion corrected image generation unit 206 and corrected for radial distortion.

  The display unit 402 outputs information about the obtained object such as the type and distance of the object recognized by the image recognition unit 401. The display unit 402 displays, for example, information on the object obtained by performing processing for emphasizing the object on an image in which the object is recognized or adding information indicating the distance on the screen. Alternatively, the user may be notified by voice.

  If the captured image has a radial distortion, it may be difficult to recognize a predetermined object such as a pedestrian, an obstacle, a vehicle, or a road sign. Since the distortion is corrected, the original recognition performance can be realized.

  FIG. 13 is a flowchart showing an example of the operation of the surrounding environment recognition apparatus of the present embodiment. In addition, since the process of step S301-S307 is the same as the process of step S201-S207 mentioned above, description is abbreviate | omitted. Although not shown in the figure, before step S307, the time series determination unit 3052 obtains the median from the distortion correction parameter group stored in the distortion correction parameter storage unit 3051, and determines the final distortion correction parameter. You may decide.

  When the distortion-corrected image is generated in step S307, the image recognition unit 401 performs image recognition on the distortion-corrected image, and when the target object is recognized, obtains information on the recognized target object from the image (step S307). S308).

  Next, the display unit 402 outputs information obtained as a result of the image recognition process (step S309).

  By repeating the above operation at a specified timing or while the imaging unit 201 is activated, image recognition is performed on an image whose distortion has been corrected by the calculated correction parameter value while always obtaining the latest distortion correction parameter. be able to.

  FIG. 14 is a block diagram showing another configuration example of the surrounding environment recognition apparatus of this embodiment. If it is considered that the distortion correction parameter does not change for a certain period, as shown in FIG. 14, the driving system of the surrounding environment recognition apparatus, the processing system for estimating the distortion correction parameter, and the distortion correction by the estimated distortion correction parameter are performed. It is also possible to adopt a configuration in which only one of the processing systems is driven at a single time, separated from the processing system that performs image recognition on the image subjected to. The surrounding environment recognition apparatus illustrated in FIG. 14 further includes a control switching unit 403 as compared with the configuration illustrated in FIG.

  That is, in the configuration shown in FIG. 14, a processing system for estimating distortion correction parameters (temporary distortion correction means 202, local gradient direction histogram calculation means 203, entropy calculation means 204, and distortion correction is provided after the imaging means 201. A system including the parameter determination unit 305 (hereinafter referred to as a first system) and a processing system that performs image recognition using the estimated distortion correction parameters (distortion correction image generation unit 206, image recognition unit 401, and display unit 402). (Hereinafter referred to as a second system) is separately connected via the control switching means 403. In this example, apart from the distortion correction parameter storage means 3051 for time series processing, a fixed distortion correction parameter storage means (which stores a distortion correction parameter that is currently valid (referred to as a fixed distortion correction parameter)). (Not shown).

  The control switching unit 403 switches the system to be operated in accordance with the determination status of the distortion correction parameter.

  In the present embodiment, the distortion correction image generation unit 206 uses the deterministic distortion correction parameter stored in the deterministic distortion correction parameter storage unit described above with respect to the captured image input from the imaging unit 201. Is generated.

  FIG. 15 is a flowchart showing another example of the operation of the surrounding environment recognition apparatus having the configuration shown in FIG.

  In the example illustrated in FIG. 15, first, the imaging unit 201 captures a captured image and captures a scene or the like reflected in the installed state (step S401).

  Next, the control switching unit 403 determines whether or not the distortion correction parameter has been confirmed, and switches the processing system to be operated according to the determination result (step S402). The control switching unit 403, for example, determines that the distortion correction parameter has been determined when the valid distortion correction parameter storage unit stores a valid value or that it is valid, and stores an invalid value or invalidity. If it is, it may be determined as indeterminate.

  If the distortion correction parameter has not been determined, the control switching unit 403 switches the operating system to the first system and advances the process to step S403 (No in step S402).

  In step S403, the temporary distortion correction unit 202, the local gradient direction histogram calculation unit 203, the entropy calculation unit 204, and the distortion correction parameter determination unit 305 perform distortion correction parameter estimation processing on the captured image input in step S401. I do. Specifically, the processes in steps S302 to S306 described above are repeated until there is no unknown parameter, and the distortion correction parameter that most corrects the distortion for the captured image at the current time is estimated.

  Next, the time series determination unit 3051 of the distortion correction parameter determination unit 305 attempts to estimate the time series of the distortion correction parameter using the past distortion correction parameter estimation result (step S404). The time series determination unit 3051 determines the estimation result obtained when the estimation results of the past sufficient number of distortion correction parameters are sufficient for the time series estimation and the estimation result is estimated to be valid. The distortion correction parameter is stored in the distortion correction parameter storage means, and the distortion correction parameter is determined to be confirmed (Yes in step S405, step S406). In other cases, the distortion correction parameter remains unconfirmed (No in step S405). Then, the process returns to step S401.

  On the other hand, when the distortion correction parameter has been determined in step S402, the control switching unit 403 switches the system to be operated to the second system and advances the process to step S407 (Yes in step S402).

  In step S407, the distortion correction image generation unit 206 performs conversion processing on the input captured image using the definite distortion correction parameter, and generates a distortion correction image.

  Next, the image recognition unit 401 performs image recognition processing on the generated distortion-corrected image (step S408).

  Next, the display unit 402 displays and notifies the recognition result (step S409).

  Finally, the control switching unit 403 determines whether or not it is the update timing of the distortion correction parameter, and if it is the update timing (Yes in step S410), the definite distortion correction parameter stored in the definite distortion correction parameter storage unit is set. Invalidate and return to step S401 (step S411). If it is not the update timing, the process directly returns to step S401 and waits for the next image to be input. The update timing is not particularly limited. For example, the control switching unit 403 may invalidate the distortion correction parameter every time processing for one frame of the image is completed.

  As described above, according to this embodiment, after the distortion correction parameter is once determined, only the second system (processing system for performing image recognition) needs to be executed until it becomes invalid again. Since the estimation process is not performed, the calculation amount can be reduced.

  FIGS. 12 and 14 show examples of estimating distortion correction parameters using the configuration of the radial distortion correction apparatus shown in the third embodiment. Specific examples of distortion correction parameter estimation processing are shown in FIGS. The typical method is not limited to the method using the radial distortion correction apparatus of the third embodiment. That is, the method using the radial distortion correction apparatus described in each of the above embodiments can be used.

  Next, the outline of the present invention will be described. FIG. 16 is a block diagram showing an outline of the radial distortion correction apparatus of the present invention. The radial distortion correction apparatus shown in FIG. 16 includes an image input unit 11, a temporary distortion correction unit 12, a local gradient direction histogram calculation unit 13, a bias calculation unit 14, and a distortion correction image selection unit 15. .

  The image input unit 11 (for example, the interface unit of the imaging unit 101) inputs a captured image taken by an imaging device that is a device for correcting radial distortion.

  The temporary distortion correction unit 12 (for example, the temporary distortion correction unit 102 and the temporary distortion correction unit 202) assumes two or more distortion correction parameters for the captured image input by the image input unit 11, and A temporary distortion correction image corresponding to each assumed distortion correction parameter is generated.

  The local gradient direction histogram calculation unit 13 (for example, the local gradient direction histogram calculation unit 103 and the local gradient direction histogram calculation unit 203) performs image processing on each of the temporary distortion correction images generated by the temporary distortion correction unit 12. A local gradient direction histogram representing a distribution of local gradient directions, which is a gradient direction in a local region, is calculated.

  The bias calculation means 14 (for example, the entropy calculation means 104 and the entropy calculation means 204), based on the local gradient direction histogram calculated by the local gradient direction histogram calculation means 13, determines the local gradient direction in each temporary distortion correction image. Calculate the degree of bias.

  The distortion correction image selection means 15 (for example, the distortion correction image selection means 105, the parameter sequential determination means 205) is based on the degree of bias in the local gradient direction in each of the temporary distortion correction images calculated by the bias calculation means 14. Then, one distortion correction image is selected from the temporary distortion correction images.

  According to such a configuration, distortion correction parameters can be automatically optimized from an input captured image without using a special calibration device in which lattice points are regularly arranged like a checkerboard. An image in which the radial distortion of the camera is suitably corrected can be automatically obtained.

  Further, the distortion correction image selection unit 15 may select a temporary distortion correction image having the largest degree of bias in the local gradient direction as the distortion correction image.

  Further, the radial distortion correction device is a distortion correction parameter determination unit that determines a distortion correction parameter to be applied to a captured image input after the current time, based on a selection result of the distortion correction image by the distortion correction image selection unit. For example, a distortion correction parameter determination unit 305) may be provided.

  Further, the provisional distortion correction unit 12 selects one unknown parameter included in the distortion correction parameter, sets a plurality of values of the selected unknown parameter, and assumes two or more distortion correction parameters. Good. In such a case, the distortion correction parameter determination unit uses, as the distortion correction parameter, processing for determining the value of the unknown parameter selected by the temporary distortion correction unit 12 based on the selection result of the distortion correction image selection unit. This may be done sequentially until the values of all the unknown parameters involved are determined.

  Further, the temporary distortion correction unit 12 may sequentially select parameters corresponding to lower-order terms among unknown parameters included in the distortion correction parameters in the distortion correction model expression expressed by a polynomial.

  Further, the radial distortion correction apparatus stores, as an estimated distortion correction parameter, a distortion correction parameter used to generate a temporary distortion correction image selected as the distortion correction image for a captured image that is input every moment. Distortion correction parameter storage means is provided, and the distortion correction parameter determination means calculates a median from the distortion correction parameter group estimated for the most recent predetermined number of captured images stored in the distortion correction parameter storage means. The median value of the calculated distortion correction parameters may be used as a distortion correction parameter to be applied to images input after the current time.

  Further, the local gradient direction histogram calculation means 13 is orthogonal to the local gradient direction obtained from the area around any pixel in the image and the magnitude of the gradient and the local gradient direction in the area for which the local gradient direction is obtained. The local gradient direction histogram may be calculated using a value corresponding to the distance between the straight line and the center of the image as a weight.

  Although the present invention has been described with reference to the exemplary embodiments and examples, the present invention is not limited to the above exemplary embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  Moreover, although a part or all of said embodiment can be described also as the following additional remarks, it is not restricted to the following.

(Appendix 1)
Assuming that an image input unit that inputs a captured image captured by an imaging device that is a correction target device for radial distortion and two or more distortion correction parameters are assumed for the captured image input by the image input unit. A temporary distortion correction unit that generates a temporary distortion correction image corresponding to each distortion correction parameter, and a temporary distortion correction image generated by the temporary distortion correction unit, in a local region in the image. A local gradient direction histogram calculating means for calculating a local gradient direction histogram representing the distribution of the local gradient direction, which is a gradient direction, and a local gradient direction histogram calculated by the local gradient direction histogram calculating means, Bias calculation means for calculating the degree of bias in the local gradient direction in each, and provisional distortion compensation calculated by the bias calculation means Based on the degree of local gradient direction of bias in each image, the radiation distortion correction apparatus characterized by comprising a distortion correction image selection means for selecting one of the distortion correction image from the temporary distortion correction image.

(Appendix 2)
The bias calculation means calculates the entropy of the local gradient direction histogram for each temporary distortion correction image as an index indicating the degree of bias in the local gradient direction, and the distortion correction image selection means gives the minimum value of entropy. The radial distortion correction apparatus according to appendix 1, wherein a temporary distortion correction image is selected as a distortion correction image.

(Appendix 3)
The bias calculation means calculates the variance, standard deviation or sum of squares of the local gradient direction histogram for each of the temporary distortion correction images as an index indicating the degree of bias in the local gradient direction, and the distortion correction image selection means The radial distortion correction apparatus according to appendix 1, wherein a temporary distortion correction image that gives a maximum value of variance, standard deviation, or sum of squares is selected as a distortion correction image.

(Appendix 4)
A distortion correction parameter for storing a predetermined number of temporary distortion correction parameters used to generate a temporary distortion correction image selected as a distortion correction image as a distortion correction parameter of an estimation result for a captured image input every moment Storage means, and the distortion correction parameter determination means is the latest obtained by adding the distortion correction parameter of the estimation result obtained for the captured image at the current time to the distortion correction parameter of the estimation result stored in the distortion correction parameter storage means. The median value is calculated from the distortion correction parameter group of the estimation results, and the calculated median value is used as a distortion correction parameter to be applied to images input after the current time. If the index indicating the degree of bias in the image group does not take an extreme value, or if the extreme value is not clear, a distortion-corrected image is selected. Results have been provisionally radiation distortion correction apparatus according to the distortion correction parameter to be excluded note from the storage object 1 used to generate the distortion correction image.

(Appendix 5)
An image input unit that inputs a captured image captured by an imaging device installed in a vehicle, and two or more distortion correction parameters are assumed for the captured image input by the image input unit, and each distortion correction is assumed. A temporary distortion correction unit that generates a temporary distortion correction image corresponding to the parameter, and a temporary distortion correction image generated by the temporary distortion correction unit, in a gradient direction in a local region in the image. A local gradient direction histogram calculating means for calculating a local gradient direction histogram representing a distribution of a certain local gradient direction, and a local gradient direction histogram in each of the provisional distortion correction images based on the local gradient direction histogram calculated by the local gradient direction histogram calculation means; Each of bias calculation means for calculating the degree of bias in the gradient direction, and provisional distortion correction images calculated by the bias calculation means Based on the degree of bias in the local gradient direction, distortion correction image selection means for selecting one distortion correction image from the temporary distortion correction images, and distortion correction image selection results by the distortion correction image selection means. A distortion correction parameter determining unit that determines a distortion correction parameter to be applied to a captured image that is input after the current time, and an image that is input with the distortion correction parameter determined by the distortion correction parameter determining unit being input after the current time. An image that is applied to an image to generate a distortion-corrected image and an image that recognizes a road environment around the vehicle by performing image recognition processing on the distortion-corrected image generated by the distortion-corrected image generating unit. A road environment recognition apparatus comprising: a recognition means.

(Appendix 6)
The road environment recognition apparatus according to appendix 5, wherein the image recognition means executes at least one of pedestrian detection, obstacle detection, road sign recognition, and vehicle detection using an image recognition technique.

(Appendix 7)
The road environment recognition apparatus according to appendix 5 or appendix 6, comprising display means for displaying an image recognition result by the image recognition means.

  The present invention is not limited to the application for correcting the radial distortion of the in-vehicle camera, and can be suitably applied as long as it is an apparatus, system, method, and program that use a captured image captured by an imaging device that may generate radial distortion.

11 Image input means 101, 201 Imaging means 12, 102, 202 Temporary distortion correction means 13, 103, 203 Local gradient direction histogram calculation means 14 Bias calculation means 104, 204 Entropy calculation means 15, 105 Distortion correction image selection means 205 Parameters Sequential determination means 206 Distortion correction image generation means 305 Distortion correction parameter determination means 3051 Distortion correction parameter storage means 3052 Time series determination means 401 Image recognition means 402 Display means 403 Control switching means 601 ECU
602 Camera 603 CPU
604 RAM
605 ROM
606 Signal processing circuit 607 Power supply circuit

Claims (10)

Image input means for inputting a captured image taken by an imaging device which is a device to be corrected for radial distortion;
Temporary distortion correction means for generating a temporary distortion correction image corresponding to each assumed distortion correction parameter assuming two or more distortion correction parameters for the captured image input by the image input means;
For each of the temporary distortion correction images generated by the temporary distortion correction means, a local gradient direction that calculates a local gradient direction histogram representing a distribution of local gradient directions that are gradient directions in local regions in the image. Histogram calculation means;
Bias calculation means for calculating the degree of bias in the local gradient direction in each of the temporary distortion correction images based on the local gradient direction histogram calculated by the local gradient direction histogram calculation means;
Distortion correction image selection means for selecting one distortion correction image from the temporary distortion correction images based on the degree of bias in the local gradient direction in each of the temporary distortion correction images calculated by the bias calculation means; A radial distortion correction apparatus comprising: The radial distortion correction apparatus according to claim 1, wherein the distortion correction image selection unit selects a temporary distortion correction image having the largest degree of bias in the local gradient direction as a distortion correction image. The distortion correction parameter determination unit that determines a distortion correction parameter to be applied to a captured image input after the current time based on a selection result of the distortion correction image by the distortion correction image selection unit. The radial distortion correction apparatus according to 2. The temporary distortion correction means selects one unknown parameter included in the distortion correction parameter, sets a plurality of values of the selected unknown parameter, assumes two or more distortion correction parameters,
The distortion correction parameter determination unit performs a process of determining the value of the unknown parameter selected by the temporary distortion correction unit based on the selection result of the distortion correction image selection unit. The radial distortion correction apparatus according to claim 3, which is sequentially performed until a parameter value is determined. 5. The radiation according to claim 4, wherein the temporary distortion correction means selects in order from a parameter corresponding to a lower-order term among unknown parameters included in a distortion correction parameter in a distortion correction model expression expressed by a polynomial. Distortion correction device. Distortion correction parameter storage means for storing, as an estimated distortion correction parameter, a distortion correction parameter used to generate a temporary distortion correction image selected as a distortion correction image for a captured image input every moment.
The distortion correction parameter determination unit calculates a median from the distortion correction parameter group estimated for the most recent predetermined number of captured images stored in the distortion correction parameter storage unit, and calculates the calculated distortion correction parameter. The radial distortion correction apparatus according to any one of claims 3 to 4, wherein a median value of is used as a distortion correction parameter to be applied to an image input after the current time. The local gradient direction histogram calculation means calculates the magnitude of the gradient with respect to the local gradient direction obtained from an area around any pixel in the image, and a straight line orthogonal to the local gradient direction in the area for which the local gradient direction is obtained. The radial distortion correction apparatus according to any one of claims 1 to 6, wherein a local gradient direction histogram is calculated using a value according to a distance between the image center and a distance as a weight. An image input means for inputting a captured image captured by an imaging device installed in the vehicle;
Temporary distortion correction means for generating a temporary distortion correction image corresponding to each assumed distortion correction parameter assuming two or more distortion correction parameters for the captured image input by the image input means;
For each of the temporary distortion correction images generated by the temporary distortion correction means, a local gradient direction that calculates a local gradient direction histogram representing a distribution of local gradient directions that are gradient directions in local regions in the image. Histogram calculation means;
Bias calculation means for calculating the degree of bias in the local gradient direction in each of the temporary distortion correction images based on the local gradient direction histogram calculated by the local gradient direction histogram calculation means;
Distortion correction image selection means for selecting one distortion correction image from the temporary distortion correction images based on the degree of bias in the local gradient direction in each of the temporary distortion correction images calculated by the bias calculation means; ,
Distortion correction parameter determination means for determining a distortion correction parameter to be applied to a captured image input after the current time based on a distortion correction image selection result by the distortion correction image selection means;
Applying a distortion correction parameter determined by the distortion correction parameter determination unit to a captured image input after the current time to generate a distortion correction image generation unit;
A road environment recognition apparatus comprising: an image recognition unit that performs image recognition processing on the distortion correction image generated by the distortion correction image generation unit and recognizes a road environment around the vehicle. Input a captured image taken by an imaging device that is a device for correcting radial distortion,
Assuming two or more distortion correction parameters for the input captured image, a temporary distortion correction image corresponding to each assumed distortion correction parameter is generated,
For each generated temporary distortion corrected image, calculate a local gradient direction histogram representing the distribution of the local gradient direction, which is the gradient direction in a local region in the image,
Based on the calculated local gradient direction histogram, calculate the degree of local gradient direction bias in each of the provisional distortion correction image,
A radial distortion correction method, wherein one distortion correction image is selected from the temporary distortion correction images based on a degree of local gradient direction bias in each of the calculated temporary distortion correction images. On the computer,
An image input process for inputting a captured image taken by an imaging device that is a target device for correcting radial distortion,
Temporary distortion correction processing for generating a temporary distortion correction image corresponding to each assumed distortion correction parameter assuming two or more distortion correction parameters for the input captured image;
A local gradient direction histogram calculation process for calculating a local gradient direction histogram representing a distribution of local gradient directions, which is a gradient direction in a local region in the image, for each of the generated temporary distortion correction images;
A bias calculation process for calculating a degree of bias in the local gradient direction in each of the temporary distortion correction images based on the calculated local gradient direction histogram, and a local gradient in each of the calculated temporary distortion correction images A radial distortion correction program for executing distortion correction image selection processing for selecting one distortion correction image from the temporary distortion correction images based on the degree of direction bias.

Images