JP4285618B2 - Stereo camera self-diagnosis device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a self-diagnosis device for a stereo camera that performs self-diagnosis of a mechanical position shift between cameras in a stereo camera and corrects the mechanical position shift by image conversion.
[0002]
[Prior art]
In general, as a three-dimensional measurement technique based on images, a correlation between a pair of images obtained by capturing an object from different positions with two cameras (stereo cameras) is obtained, and the stereo camera mounting position, focal length, and the like from the parallax for the same object Image processing by a so-called stereo method is known in which a distance is obtained by the principle of triangulation using the camera parameters.
[0003]
In this stereo image processing, two image signals obtained from a stereo camera are sequentially shifted and overlapped to obtain a position where the two image signals coincide with each other. It is desirable that there is no deviation, and it is important to adjust the optical position of the stereo camera.
[0004]
For this reason, the present applicant has previously found a matching point by processing an image of a known distance adjustment pattern imaged by a stereo camera to be adjusted in Japanese Patent Laid-Open No. 10-115506. A technique for precisely adjusting the positional displacement of the stereo camera via an actuator by calculating the translational displacement amount and the rotational displacement amount of the image by simple calculation. Also, in Japanese Patent Application No. 10-61578, The main camera is mechanically rotated around the optical axis so that the horizontal line of the main camera is parallel to the baseline of both cameras, and the sub camera image is affine transformed with the main camera image as a reference to adjust as a stereo camera We are proposing a technology to do this.
[0005]
[Problems to be solved by the invention]
However, stereo cameras that have been adjusted at the time of assembly and shipment are inevitably affected by changes in the adjustment position due to secular changes and vibration shocks during operation in the actual field of use. The capability is gradually reduced, the accuracy of distance information obtained by stereo processing of the captured image is deteriorated, and the reliability of the entire system is reduced.
[0006]
The present invention has been made in view of the above circumstances, and provides a self-diagnosis device for a stereo camera that can self-diagnose a shift of an adjustment position of a stereo camera and automatically correct the position shift while being incorporated in a system. The purpose is to do.
[0007]
[Means for Solving the Problems]
  In order to achieve the above object, the invention according to claim 1 is a self-diagnosis device for a stereo camera that self-diagnose a mechanical position shift between cameras in a stereo camera and corrects the mechanical position shift by image conversion. Images taken with the stereo cameraInsideInA region included in a region that is a predetermined region and is a candidate for processing for calculating a shift of a corresponding position between a pair of original images to be subjected to stereo processingVariance of brightness changeIs greater than the predetermined lower threshold, A process for calculating a shift of a corresponding position between a pair of original images to be subjected to stereo processingAllow executionFor calculating the shift of the corresponding position between the means and the original imageAllowed to runDispersion of the brightness change amountSo that the area is larger than the lower thresholdMeans for calculating a deviation calculation area in the original image, calculating a deviation distribution of corresponding positions between the areas, and calculating a coordinate conversion value of an image captured by the stereo camera from the deviation distribution. And a means for correcting the mechanical positional deviation of the stereo camera based on the converted value.
[0008]
  The invention according to claim 2 is the invention according to claim 1,Comprising shutter control means for controlling the shutter speed of the stereo camera;The range for calculating the above deviationShutter speed by the shutter control meansThe brightness extraction range for control is made to coincide.
[0009]
In the invention of claim 3, in the invention of claim 1, when the deviation distribution is linearly approximated by the Hough transform, the voting surface of the Hough transform is gradually expanded from coarse to dense by re-dividing in the most votes range. It is characterized by that.
[0010]
  That is, in the first aspect of the invention,, SuImages taken with a teleo cameraInsideInA region included in a region that is a predetermined region and is a candidate for processing for calculating a shift of a corresponding position between a pair of original images to be subjected to stereo processingVariance of brightness changeIs greater than the predetermined lower threshold, A process for calculating a shift of a corresponding position between a pair of original images to be subjected to stereo processingIn the original image so that the region where the variance of the brightness change amount is larger than the lower threshold is included.A shift calculation area is set, and a shift distribution of corresponding positions between the areas is calculated. Then, a coordinate conversion value of an image captured by the stereo camera is calculated from the distribution of the shift, and an image captured by the stereo camera is converted using the coordinate conversion value to correct a mechanical position shift of the stereo camera.
[0011]
  In this case, as described in claim 2,Equipped with shutter control means for controlling the shutter speed of the stereo camera,The range to calculate the deviationShutter speed by shutter control meansIt is desirable to match the luminance extraction range for control, and an optimal exposure for deviation detection can be obtained.
[0012]
In addition, as described in claim 3, when the deviation distribution is linearly approximated by the Hough transform, it is desirable that the voting surface of the Hough transform is expanded in order from coarse to dense by subdividing in the most frequent voting range. The amount of memory required for the voting surface for conversion can be reduced.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described below with reference to the drawings. 1 to 7 relate to an embodiment of the present invention, FIG. 1 is a flowchart of a camera shift self-diagnosis routine, FIG. 2 is a block diagram of an image processing system related to a self-diagnosis of a stereo camera, and FIG. FIG. 4 is an explanatory diagram showing a luminance extraction range of shutter control, FIG. 5 is an explanatory diagram showing a city block calculation region, and FIG. 6 is an explanatory diagram showing a Hough transform voting plane. 7 is an explanatory diagram of a deviation correction amount.
[0014]
In FIG. 2, reference numeral 1 denotes a stereo camera composed of a set of two cameras 2 and 3 that are synchronized with each other and have a variable shutter speed, and captures a reference image when one camera 2 performs stereo processing. As a sub camera that captures a comparison image when the main camera and the other camera 3 perform stereo processing, for example, a three-dimensional distance obtained from a pair of images that are mounted on a moving body such as an automobile or a helicopter and that captures a target landscape It is adopted in a system that performs processing for recognizing surrounding environment and self-location based on information.
[0015]
The stereo camera 1 is connected to a stereo image processing device 30 via an image signal processing unit 10 that performs input processing of imaging signals. The image signal processing unit 10 corresponds to each system of the cameras 2 and 3. Analog interfaces 11 and 12 for adjusting the gain / offset of the analog imaging signals from the cameras 2 and 3 in accordance with the input range of the subsequent stage, the analog image with a predetermined luminance gradation (for example, a gray scale of 256 gradations) A / D converters 13, 14, etc. for converting to a digital image) are provided, and the sub-camera 3 side stores the digitized comparison image temporarily in the input image memory 15 and the input image memory 15. An affine transformation circuit 16 is provided for performing geometric transformation such as image rotation and translation on the obtained image.
[0016]
The stereo image processing device 30 includes various circuits for stereo processing, such as an adder, a difference unit, and an absolute value arithmetic circuit, a memory, and the like. A distance image obtained by performing stereo matching processing to identify the corresponding small areas by calculating the city block distance and calculating the correlation between them, and imaging the pixel shift due to parallax according to the distance to the object as distance data Is generated.
[0017]
In the present embodiment, since the image signal processing unit 10 includes the affine transformation circuit 16 only in the system on the sub camera 3 side, the pair of original images to be subjected to stereo processing is the reference image captured by the main camera 2. FIG. 6 is an image obtained by affine transformation of a comparison image captured by the sub camera 3. However, both the main camera 2 side and the sub camera 3 side may be provided with an affine transformation circuit. In this case, a pair of original images to be subjected to stereo processing is a reference taken by the main camera 2. An image obtained by affine transformation of an image obtained by affine transformation of the image and a comparison image captured by the sub camera 3 is obtained.
[0018]
Here, the stereo camera 1 is attached to a camera stay (not shown) with a predetermined baseline length, and the positional relationship between the cameras 2 and 3 is adjusted and fixed at the time of assembly and shipment. Regarding the assembly adjustment at the time of shipment of the stereo camera 1, the applicant previously disclosed in Japanese Patent Application No. 10-61578 that the main camera is rotated around the optical axis so that the horizontal line of the main camera is parallel to the base line of both cameras. We have proposed a technology to adjust the rotation as a stereo camera by affine transformation of the sub camera image based on the image of the main camera as a reference. By adjustment, the stereo camera 1 can be adjusted at the time of shipment.
[0019]
However, the two cameras 2 and 3 constituting the stereo camera 1 are mechanically misaligned with respect to the state at the time of assembly and shipment due to changes over time, vibration and shock during operation, and the like. For this reason, in the actual field operating state, the adaptability to the texture of the surface to be imaged and the accuracy of the distance information gradually deteriorate, and mechanical position adjustment and electrical image correction are used in combination. Even when assembly adjustment is performed, the affine transformation parameters set at the time of shipment do not match the actual situation. In any case, the adjustment position of the stereo camera 1 is inevitably deviated from the initial state due to a change with time regardless of the adjustment at the time of assembly and shipment, which causes a decrease in the reliability of the entire system.
[0020]
For this reason, the image signal processing unit 10 performs self-diagnosis of the mechanical position shift of the stereo camera 1 immediately before the system is actually operated immediately after the power is turned on, and electrically corrects the slight shift of the mechanical optical position. Accordingly, a self-diagnosis unit 20 for correcting / updating the conversion parameter of the affine conversion circuit 16 is connected.
[0021]
The self-diagnosis unit 20 includes image memories 21 and 22 that store the images from the A / D converters 13 and 14, an image memory 23 that stores an output image from the affine transformation circuit 16, and these image memories 21, 22 and 23. It is composed of a CPU 24 and the like that self-diagnose the positional deviation of the stereo camera 1 based on the image stored in the image and corrects / updates the conversion parameter of the affine conversion circuit 16. In an altitude measurement system that measures the altitude by stereo-processing images taken from the ground, such as helicopters, self-diagnosis is performed using a still image obtained by imaging the ground while the aircraft is warming up before takeoff immediately after the system power is turned on. . The self-diagnosis unit 20 may be configured as a part of the stereo image processing device 30 or the recognition processing device.
[0022]
The self-diagnosis process in the self-diagnosis unit 20 includes a first algorithm for determining whether or not the position shift calculation of the stereo camera 1 can be performed, a second algorithm for calculating a shift amount (city block distance), and a calculation The third algorithm is used to evaluate the amount of deviation and correct the affine transformation parameters. In the first algorithm, the exposure amount is abnormal, the signal system is disconnected, or the brightness of the imaged subject is uniform. For this reason, when there is little change in luminance in the image captured by the stereo camera 1, even if the positional deviation is calculated, the probability that erroneous detection will occur becomes extremely high and the reliability is reduced. Whether or not the situation is surely possible is determined using the variance of the luminance change amount in the captured image.
[0023]
In addition, in the second algorithm, when it is determined that the shift calculation can be performed by the first algorithm, even if an image with extremely different luminance is reflected in a range outside the shift detection target in the image, it is optimal for the shift detection region. The city block distance between the reference area and the comparison area is calculated after optimizing the shutter control so that a correct exposure can be obtained, and further, sub-pixel interpolation is performed to obtain a deviation amount of one pixel or less.
[0024]
In the third algorithm, a temporary profile of the deviation line is obtained from the obtained deviation amount of each region by Hough transform, and further, the slope and intercept of the deviation line are determined by the least square method. Then, the rotation amount and translation amount of the shift are calculated from the slope and intercept of the shift straight line, and if they are appropriate, they are written in the register of the affine transformation circuit 16 to correct the shift adjustment.
[0025]
Specifically, the above self-diagnosis process is performed by a camera shift self-diagnosis routine shown in FIG. Hereinafter, the camera shift self-diagnosis routine will be described.
[0026]
The camera misalignment self-diagnosis routine is a routine that is first executed after the system is turned on and initialized, and before the system enters an actual operation state. First, in step S101, an image range suitable for misalignment detection is used. In order to select, the image captured by the stereo camera 1 in a stationary state before the system is actually operated is read from the image memories 21 and 22, and the variance of the luminance change amount in the image (hereinafter, luminance variance) σ²Search for the row jmax that maximizes.
[0027]
That is, as shown in FIG. 3, a plurality of strip-shaped deviation detection candidate areas are set in the image, and these deviation detection candidate areas and the luminance variance σ are set.²When attention is paid to one region N among a plurality of regions where the row j to be obtained overlaps, the luminance dispersion σ possessed by this region N² _NUses the average value bave (= (bi-1 + bi + bi + 1) / 3) when the luminance of the central coordinate i of the region N is bi and the luminances of the left and right neighbors are bi-1, bi + 1, respectively. (1).
σ² _N= (│bi-1-bave│ + │bi-bave│ + │bi + 1-bave│)²... (1)
[0028]
Since the deviation detection candidate areas are set over a plurality of columns in the image, if the luminance variance is calculated in the same manner for other overlapping areas, the luminance variance σ of the focused row j is eventually obtained.²(j) is the luminance variance σ of each region as shown in the following equation (2).² _NIs given as the sum of
σ²(j) = Σσ² _N    ... (2)
[0029]
Then, the calculation of the sum of the formula (2) is performed for all the rows in the range overlapping with the deviation detection candidate region, and the luminance variance of each row is multiplied by a coefficient K (j) weighted at the center of the image. The coefficient K (j) is a coefficient for removing the influence of lens distortion using the data at the center of the image, and the maximum line in the distribution of luminance dispersion weighted by the coefficient K (j) is represented by jmax. And
[0030]
Next, the process proceeds to step S102, and the obtained luminance variance σ²Is compared with the lower threshold. This lower limit threshold value is obtained in advance by simulation or experiment to obtain a lower limit value of luminance dispersion that can detect displacement effectively. And luminance variance σ²Is less than the lower limit threshold, it is determined that the reliability of deviation detection is low, and the process exits the routine from step S102, and the luminance dispersion σ²If it exceeds the lower limit threshold, it is determined that deviation detection can be performed, and the process proceeds to step S103.
[0031]
In step S103, as shown in FIG. 4, the brightness extraction range for shutter control is limited from the entire normal image to the deviation detection candidate area, and the shutter time is set so that optimal exposure can be obtained in the deviation detection candidate area. In other words, the area where displacement detection is performed is a limited area in the image, and normal shutter control for the entire image may not provide an appropriate shutter speed due to unnecessary reflections or shadows. . For this reason, the luminance dispersion σ²Is limited to a region having a predetermined width above and below the maximum row jmax, and this luminance dispersion σ²By performing luminance extraction only on the region determined by the above and determining the shutter speed, it is possible to obtain the optimum luminance for displacement detection.
[0032]
In subsequent step S104, the stereo camera 1 captures again a set of reference images and comparison images used for displacement detection at the shutter speed set in step S103. In step 105, the reference image captured by the main camera 2 and the image obtained by affine transformation of the comparison image captured by the sub camera 3 are read from the image memories 21 and 23, respectively, and these images are used as the original images and shifted. An area for calculating the city block distance is set from the detection candidate area, and deviation detection and sub-pixel interpolation are performed for each area.
[0033]
When the assembly adjustment of the stereo camera 1 is performed only by mechanical adjustment and diagnosis is started for the first time, the image captured by the sub camera 3 is passed through without being transformed by the affine transformation circuit 16, and the comparison side The original image. Further, when an affine transformation parameter for an image captured by the sub camera 3 is written in the register of the affine transformation circuit 16 at the time of assembly adjustment, or when a diagnosis has been performed last time, the coordinates are determined according to the current affine transformation parameter. The converted image becomes the comparison-side original image.
[0034]
That is, first, as shown in FIG. 5, the reference area in the original image on the reference side is set around the line jmax having the maximum luminance variance for the deviation detection candidate area, and the necessary search range is set from the reference area. The comparison area in the original image on the comparison side is set at a position that is larger in size and shifted in the horizontal direction (i direction) by the parallax D calculated backward from the distance to the imaging target.
[0035]
Next, while changing the comparison start coordinates between the reference region and the comparison region that form a pair, as shown in the following equation (3), the luminance data Ki, j of each pixel in the reference region and the comparison region The city block distance C is calculated by summing the absolute values of the differences from the luminance data Hi, j of each pixel, and the coordinates at which the city block distance C is minimum are obtained as temporary matching points (iM, jM).
C = Σ│Ki, j-Hi, j│ (3)
[0036]
Furthermore, based on the premise that the distribution of the city block distance in the vicinity of the temporary coincidence point (iM, jM) is symmetrical around the local minimum point, from the magnitude relation of the difference in the city block distance before and after the temporary coincidence point The position of the minimum point is specified, and the subpixel interpolation for calculating the coordinates of the minimum point by linear approximation is performed to estimate the matching point (iS, jS) of the subpixel with a resolution of one pixel or less.
[0037]
Then, as shown in the following equation (4), the difference between the coordinate point (i0, j0) representing the region and the coincidence point (iS, jS) of the subpixel is expressed as the amount of deviation j of the corresponding region.^*Get as. This process is performed for all regions, and the deviation amount j with respect to the coordinate i0^*Get the distribution of. When the city block distance value is larger than a preset value (when the degree of coincidence is low), the detection result of that area is invalidated.
j^*= JS-j0 (4)
[0038]
Thereafter, the process proceeds to step S106, in which it is checked whether or not the number of regions where the deviation amount distribution is obtained is equal to or greater than a predetermined number. If the number of regions from which the deviation amount distribution is obtained is smaller than the predetermined number of regions, the reliability in detecting the deviation straight line at the subsequent stage is lowered, and the routine is exited and the process is interrupted. On the other hand, if the number of areas where the deviation amount distribution is obtained is equal to or larger than the predetermined number of areas, the process proceeds to step S107, and the deviation amount j with respect to the coordinate i0^*Is a straight line j^*= Ai0 + b is assumed, a temporary deviation straight line is obtained by Hough transform, and coefficients ah (slope) and bh (intercept) are calculated.
[0039]
In this case, in the normal Hough transform, if the resolution is made fine, the amount of memory used increases. Therefore, in this embodiment, as shown in FIG. 6, initially, the number of voting planes in the parameter space is reduced to reduce the amount of memory used, and the voting plane is divided again for the range with the largest number of votes. cure. While the number of votes exceeds the default value, the most voted area is further divided and the Hough transform is repeatedly executed to expand the effective voting surface, thereby reducing the amount of memory used. The resolution can be ensured while reducing.
[0040]
When the provisional deviation straight line is obtained by the above Hough transform, the process proceeds to step S108, and a singular point remarkably different from the surroundings is removed by using the provisional deviation straight line by detecting noise or protrusions. In other words, a deviation straight line is set from the coefficients ah and bh of the temporary deviation straight line, and the deviation j for each region is set.^{* '}And the amount of deviation j^{* '}The amount of deviation j calculated from the city block distance^*If the difference between and is greater than or equal to the default value, the corresponding area is invalidated as a false detection of a matching point.
[0041]
Thereafter, the process proceeds to step S109, and it is checked whether or not the number of effective areas excluding singular points is the number of predetermined areas necessary for applying the least square method. As a result, if the number of effective areas is smaller than the predetermined number of areas, the routine is exited and the process is interrupted. If the number of effective areas is equal to or larger than the predetermined number of areas, the process proceeds to step S110.^*From these, coefficients a (slope) and b (intercept) of the deviation straight line are determined by the least square method.
[0042]
When the stereo camera 1 is shifted from the position when the adjustment is completed, the deviation line obtained by the above processing is perpendicular to the inclination with respect to the horizontal direction as shown in FIG. 7 with respect to the deviation line based on the original parallax. Resulting in a direction offset. The inclination with respect to the horizontal direction is a rotation amount to be corrected by rotating the comparison image, and the offset in the vertical direction is a translation amount to be corrected by moving the comparison image up and down.
[0043]
That is, as shown in the following formulas (5) and (6), the rotation amount Δα and the vertical translation amount Δβ applied to the comparative image can be obtained from the coefficients a and b of the deviation straight line, and the position of the stereo camera 1 can be obtained. Affine transformation parameter α to correct the deviation^*, Β^*Can be obtained by the following equations (7) and (8) using the rotation amount Δα and the vertical translation amount Δβ with respect to the affine transformation parameters α and β before diagnosis.
Δα = tan^-1a (5)
Δβ = −b (6)
α^*= Α−Δα (7)
β^*= Β−Δβ (8)
[0044]
Next, the process proceeds to step S111, and it is checked whether or not the rotation amount Δα and the vertical translation amount Δβ are within a predetermined range in which the assumed maximum mechanical deviation can be corrected and an appropriate result is obtained. As a result, if the rotation amount Δα and the vertical translation amount Δβ are not within the predetermined range, the routine is exited and the processing is interrupted. If the rotation amount Δα and the vertical translation amount Δβ are within the predetermined range, the process proceeds to step S112. Affine transformation parameter α^*, Β^*To replace the affine transformation parameters α and β before diagnosis.
[0045]
In the correction of the affine transformation parameter in step S112, an image in which the positional deviation of the stereo camera 1 is completely eliminated by one correction due to the influence of optical distortion or the aspect ratio of the coordinates of the captured image is obtained. It is not possible. Therefore, the process proceeds from step S112 to step S113, and it is checked whether or not the above process has been executed a predetermined number of times.
[0046]
This default number of times determines the affine transformation parameter as α^*→ α, β^*→ This is the number of times set in advance as the number of times that the process of replacing β can be repeated and converged to the vicinity of the true value. When the process has not been executed a predetermined number of times, the process returns to step S104, and a new reference image and A comparison image is picked up, a deviation straight line is determined by deviation detection and subpixel interpolation for each region, a rotation amount Δα and a vertical translation amount Δβ are obtained, and updating of the affine transformation parameters is continued. If the predetermined number of times has been reached, the process proceeds to step S114, where the shutter control is switched to normal control based on luminance extraction of all images, and the self-diagnosis is terminated.
[0047]
Thereafter, the system enters an operating state, and an image captured by the main camera 2 and an image obtained by converting the image captured by the sub camera 3 by the affine transformation circuit 16 are input to the stereo image processing device 30, and the stereo camera is subjected to self-diagnosis. Stereo processing is performed on the original image in which the mechanical position shift of 1 is corrected, and accurate three-dimensional distance information is generated.
[0048]
As a result, the shift of the adjustment position of the stereo camera 1 can be self-diagnosed and automatically corrected in the state of being incorporated into the system, and the position accuracy of the stereo camera 1 can be maintained at the time of shipping adjustment and the reliability of the system can be maintained. Can be improved. Moreover, in the self-diagnosis, since it is determined whether or not the subsequent shift detection is performed using the variance of the luminance change amount of the image captured by the stereo camera 1, it is possible to prevent an erroneous diagnosis and perform an accurate diagnosis. .
[0049]
【The invention's effect】
  As described above, according to the first aspect of the present invention, the image captured by the stereo cameraInsideInA region included in a region that is a predetermined region and is a candidate for processing for calculating a shift of a corresponding position between a pair of original images to be subjected to stereo processingVariance of brightness changeIs greater than the predetermined lower threshold, A process for calculating a shift of a corresponding position between a pair of original images to be subjected to stereo processingIn the original image so that the region where the variance of the brightness change amount is larger than the lower threshold is included.A shift calculation area is set, and a shift distribution of corresponding positions between the areas is calculated. Then, the coordinate conversion value of the image captured by the stereo camera is calculated from the distribution of the shift, and the image captured by the stereo camera is converted by the coordinate conversion value to correct the mechanical position shift of the stereo camera. In this state, the shift of the adjustment position of the stereo camera can be automatically corrected, and the position accuracy of the stereo camera can be maintained at the time of shipping adjustment to improve the reliability of the system.
[0050]
  Moreover, for self-diagnosis,An area where the variance of the brightness change amount is larger than the lower threshold is included.In order to set the displacement calculation area, for example, when a stereo camera is mounted on a moving body and diagnosis is performed using an image captured in a stationary state before the moving body actually operates, it is not good at detecting the displacement. It is possible to prevent an erroneous diagnosis due to the reflection of the texture and perform an accurate diagnosis.
[0051]
  In this case, in the invention according to claim 2, the range for calculating the deviation isShutter speed by shutter control meansSince the luminance extraction range for control is made to coincide, it is possible to obtain an optimum exposure for deviation detection. Further, in the invention according to claim 3, when the deviation distribution is linearly approximated by the Hough transform, the resolution is ensured because the Hough transform voting surface is enlarged gradually from the coarse to the dense by subdividing in the most frequent voting range. On the other hand, it is possible to obtain an excellent effect such as reducing the amount of memory required for the voting surface of the Hough transform.
[Brief description of the drawings]
FIG. 1 is a flowchart of a camera shift self-diagnosis routine.
FIG. 2 is a block diagram of an image processing system related to self-diagnosis of a stereo camera.
FIG. 3 is an explanatory diagram showing an area for obtaining a variance of a luminance change amount
FIG. 4 is an explanatory diagram showing a luminance extraction range of shutter control.
FIG. 5 is an explanatory diagram showing areas of city block calculation.
FIG. 6 is an explanatory diagram showing a voting plane for Hough transform
FIG. 7 is an explanatory diagram of a deviation correction amount.
[Explanation of symbols]
1 ... Stereo camera
16 ... Affine transformation circuit
20 ... Self-diagnosis department
σ² ... Dispersion of luminance change

Claims (3)

A self-diagnosis device for a stereo camera that self-diagnose a mechanical position shift between cameras in a stereo camera and corrects the mechanical position shift by image conversion,
Dispersion of the luminance variation of the region included in the candidate and a region of a process for calculating the deviation of corresponding positions between the pair of original image to be stereo processing a predetermined region in the image captured by the stereo camera Means for permitting execution of a process of calculating a shift of a corresponding position between a pair of original images to be subjected to stereo processing when is greater than a predetermined lower threshold ;
When execution of the process of calculating the shift of the corresponding position between the original images is permitted , the shift calculation region is set in the original image so that the variance of the luminance change amount is larger than the lower limit threshold. And a means for calculating a distribution of deviations of corresponding positions between the areas,
A stereo camera self comprising: means for calculating a coordinate conversion value of an image captured by the stereo camera from the distribution of the shift, and correcting a mechanical position shift of the stereo camera based on the coordinate conversion value. Diagnostic device. Comprising shutter control means for controlling the shutter speed of the stereo camera;
2. The stereo camera self-diagnosis device according to claim 1, wherein a range in which the shift is calculated and a luminance extraction range for shutter speed control by the shutter control unit are matched.   3. The stereo according to claim 1, wherein, when the deviation distribution is linearly approximated by a Hough transform, the voting surface of the Hough transform is sequentially expanded from coarse to dense by re-dividing within a most frequent voting range. Camera self-diagnosis device.

Images