KR100938195B1 - Method for distance estimation and apparatus for the same using a stereo matching - Google Patents

Abstract

PURPOSE: A method for estimating a distance using stereo matching and an apparatus thereof are provided to reduce distance estimation error while reducing complexity of distance estimation, thereby estimating the accurate location, distance and speed of a target object. CONSTITUTION: The first matching unit(720) matches a stereo image whose resolution is reduced. An object detector(730) detects an object using a depth map. The first interpolator(740) produces parallax information about the inter space of the detected object. The second matching unit matches an area of the detected object. A distance estimator(770) estimates distance to the detected object.

Description

{Method for distance estimation and apparatus for the same using a stereo matching}

The present invention relates to a distance estimating apparatus and a distance estimating method, and more particularly, to an apparatus for estimating a distance of an object included in an image by stereo matching a photographed image and a method for estimating a distance.

There are several ways to measure the distance of an object away from the observer. Among them, there is a method of capturing an image of the object and measuring a distance to an object included in the image by using the photographed image.

One example of a technique of measuring a distance to a target object included in an image by using a captured image is stereo vision technology. The stereo vision technology may measure the position, distance, and speed of an object included in an image using a left image and a right image obtained from a plurality of image capturing apparatuses, for example, cameras.

In a conventional distance measuring method using stereo vision technology, a corresponding point for an object is found from left and right images obtained from a plurality of cameras, and the distance of the object is measured using disparity with respect to the corresponding point. If the distance of the object is measured in an external environment such as a vehicle using the above method, there is a problem that a distance measurement error may occur.

In order to solve the above problems, it is necessary to increase the distance of the imaging apparatus or to increase the level of disparity. However, when the distance of the image photographing apparatus is extended, there is a problem that a shielding region may occur at a short distance. In addition, when the step of parallax is increased, the resolution of the image must be increased in order to increase the step of parallax. If the resolution of the image is increased, there is a problem of increasing complexity in matching images.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to reduce the complexity, and to provide a distance estimating apparatus and a distance estimating method capable of estimating an accurate position and distance of an object not only in a short distance but also in a long distance.

The distance estimation method according to the present invention comprises the steps of: reducing the resolution of the photographed stereo image, detecting an object by globally matching the reduced stereo image, and subsampling the internal region of the detected object. Calculating parallax information, matching the detected region of the object using the parallax information and the original stereo image, subsampling interpolation and estimating the parallax using the result of matching the object region, and Estimating a distance to the detected object using the estimated parallax.

In another aspect, an apparatus for estimating distance according to an embodiment of the present invention may include an image processor that reduces a resolution of a captured stereo image, a first matching unit that globally matches the reduced stereo image, and a depth map calculated by the first matching unit. An object detector to detect an object using a first interpolation unit to calculate parallax information by sub-sampled interpolation with respect to an internal area of the detected object, and an area of the detected object by using the parallax information and the original stereo image. A second matching unit that matches, a second interpolation unit that estimates parallax by using a result of matching by the second matching unit, and the detected object by using the parallax estimated by the second interpolation unit. It includes a distance estimator for estimating the distance of.

According to the distance estimating apparatus and the distance estimating method of the present invention, it is possible to reduce the distance estimation error while reducing the complexity of the distance estimation, and to estimate the exact position, distance, and speed of the target object not only in the distance but also in the short distance. In addition, there is an effect that can reduce the design cost of the distance estimation apparatus by reducing the distance estimation complexity.

The estimation method as described above has an effect of estimating an accurate position and distance by applying to various fields. For example, when applied to a robot or a car, it can be applied to a driving system such as collision prevention of a car, driving, or a safe driving system.

The objects, features and advantages of the present invention will become apparent from the following detailed description of embodiments taken in conjunction with the accompanying drawings. In addition, the terms used in the present invention was selected as a general term widely used as possible now, but in certain cases, the term is arbitrarily selected by the applicant, in which case the meaning is described in detail in the corresponding description of the invention, It is to be clear that the present invention is to be understood as the meaning of terms rather than names.

The operation of the distance estimating apparatus and the distance estimating method according to the present invention configured as described above will be described in detail with reference to the accompanying drawings. 1 is a diagram schematically illustrating a distance estimating method according to an exemplary embodiment of the present invention.

In the case of estimating distance using a stereo vision system, a depth map is generated by finding a corresponding point using images captured from the left and right image capturing apparatus. The moving object is detected using the depth map and the left and right images, and a parallax average in the detected moving object is obtained. In the stereo vision system, the distance of an object may be estimated using the obtained parallax. The distance between the parallax and the target object will be described in detail using Equation 1 to Equation 3 below and various parameters of FIG. 1. In FIG. 1, B (reference line) is the distance between the centers of the left and right imaging apparatuses, F is the focal length of the imaging apparatus (the distance between the lens and the imaging apparatus (eg, CCD, CMOS, etc.)), x _l is the distance from the center of the shooting device to the object on the left shooting device, x _r is the distance from the center of the shooting device to the object on the right shooting device, Z _d is the vertical distance between the object and the imaging device, and X _d is The distance from the middle point between the left image capture device and the right image capture device to the point where the horizontal line of the image pickup device meets the Z _d line, P _d is the distance between pixels of the image, and k is the number of pixels.

Since the distances D, X _d , and Z _d to the target object form a triangle, the distance D from the center position to the object can be obtained by a trigonometric formula as shown in Equation 3 below. Z _d may be obtained by dividing the distance (B) between the centers of the left and right imaging apparatuses by the focal length (F) of the imaging apparatus divided by the parallax (x _r -x _l ). Equation 1). The parallax (x _r -x _l ) can be approximated by the product of the distance P _d between the pixels and the number k of pixels. X _d may be obtained by dividing the sum of x _l and x _r by the value of Z _d obtained in Equation 1 divided by 2F (Equation 2). Therefore, in order to obtain the distance D from the object as described above, a parallax value should be obtained. Hereinafter, a process of obtaining the parallax value while reducing complexity will be described.

In the case of a stereo vision system, a depth map is obtained after matching the photographed left and right images using global matching or local matching. After detecting an object in the image based on the depth map, a distance from the detected object may be obtained by obtaining a parallax value for a center portion of the detected object. In the case of global matching, which is one of the matching methods, an accurate disparity value can be obtained with higher matching accuracy than local matching. However, when the resolution of the image is high, it takes a lot of time and calculations to match the image, and it is difficult to implement hardware for real-time processing due to the increased complexity. Therefore, first, a depth map is obtained by using a low resolution image and a global matching method, and an object is detected. The depth map is compensated for by using a high resolution image of the detected object and a local matching method to estimate the distance without increasing the complexity.

2 is a diagram schematically illustrating a process of reducing a resolution of a captured stereo image according to an embodiment of the present invention. 2 shows a left image and a right image captured by the image capturing apparatus, respectively. Each image is taken of an automobile as an object, and each has a resolution of N × M (where N and M are the number of horizontal and vertical pixels, respectively).

When the above image is captured, the resolution is reduced for each image prior to stereo image matching. For example, the resolution may be reduced by 2 ^l (l is a constant) for each image having N × M resolution. When the resolution is reduced by 2 ^l as described above, each image has a resolution of N / 2 ^l × M / 2 ^l as shown in the lower figure.

For example, when the resolution is reduced by 1/2 (l = 1), four pixels of the image (the original image) as shown in the upper part of FIG. The resolution may be reduced to correspond to one pixel. In this case, one pixel of the reduced-resolution image is equal to the sum of four pixels of the original image. The reduction may be a corresponding pixel value of an image whose resolution is reduced by using one pixel value of four pixels of the original image as a representative pixel value, and the average value of the four pixel values of the original image corresponding to the reduced image. One pixel value may be used. This may vary depending on implementation.

The resolution reduction ratio is an example, and the reduction ratio may vary depending on the embodiment, but the resolution should be sufficiently high that the object can be detected at the maximum measurable distance. As the resolution reduction ratio increases, complexity decreases, but an object detection error may increase, so the reduction ratio may be set in consideration of various factors such as accuracy requirements of a field to be applied or performance of a matching algorithm.

3 is a diagram illustrating a depth map of an image having a reduced resolution as an embodiment according to the present invention. 2, global matching may be performed on the left image and the right image by using the reduced image as shown in FIG. 2, and a depth map may be obtained. As shown in FIG. 2, a depth map of N / 2 ^l × M / 2 ^l as shown in FIG. 3 is obtained by global matching between a left image having a resolution of N / 2 ^l × M / 2 ^l and a right image. The object included in the image may be detected using the depth map obtained as described above. As a matching method for obtaining the depth map, methods such as DP (Dynamic Programming), BP (Belief Propagation), and Graph-Cut may be used.

4A and 4B are diagrams illustrating a process of detecting an object using a depth map according to an embodiment of the present invention. When the object is extracted using the depth map obtained as shown in FIG. 3, the object may be extracted using various methods, and various methods may be applied according to an implementation example. Hereinafter, for convenience of description, a case of extracting an object using a vertical disparity map (V-disparity) will be described as an example.

In the case of the photographed image, the brightness value decreases on the depth map as the distance from the photographing position gradually increases. For example, as the distance gradually increases, as shown in the left diagram of FIG. 4A, the brightness value may decrease at a constant rate on the depth map. However, when there is an object in the image, a phenomenon occurs in which the brightness value of the part where the object exists due to the object increases. Therefore, by using the above phenomenon, the portion in which the brightness value is increased can be detected as the portion having the object without decreasing the brightness value at a constant rate.

For example, a vertical parallax map like the right image of FIG. 4A may be obtained using the depth map as shown in the left image of FIG. 4A. The horizontal axis of the vertical parallax map represents a parallax value (in the case of 0 to 255, gray level 2 ⁸ ), and the vertical axis represents a vertical line of the depth map. In the case of a planar region having no object in the depth map, the brightness value decreases at a constant rate and the parallax value is proportional to the brightness value. Thus, as the distance increases, the parallax value decreases with a gentle slope as in the vertical parallax map. In the case of the region where the object is located, it has a parallax value different from the decreasing tendency. 4B is an example schematically showing a vertical parallax map obtained using the depth map as shown in FIG. 3.

When the vertical parallax map as described above is obtained, an object region is detected using the parallax map. Using the vertical parallax map as shown in FIG. 4B, a sum of regions having a disparity value and a disparity value different from each other may be detected as the object region, or an area of pixels having the same disparity value may be detected as the object region.

When the object region is detected as described above, the parallax is estimated using the internal pixels of the object region. Various methods may be used to estimate the parallax. For example, in order to estimate the parallax of the object, a sub-pixel interpolation method such as a parabola fitting method or an enhanced normalized cross correlation (ENCC) method may be used. Hereinafter, a case of using a parabola fitting method will be described as an example for convenience of description.

FIG. 5 is a diagram schematically illustrating parallax estimation according to a parabola fitting method according to an embodiment of the present invention.

The pixel closest to the center of the detected area may be used as the center pixel of the object. When the center pixel of the object is determined, one of the reduced-resolution images at the bottom of FIG. 2 is set as a reference image, and a window is set by setting a window having the center of the object as the center pixel. The reference image may be one of a left image and a right image as a reference image, which may vary depending on implementation. Hereinafter, a case in which the left image is a reference image will be described for convenience of description.

The window may have an odd size in pixels. For example, a window having a 3 × 3 size may be set as a window set as a default. This may vary depending on implementation. However, since occlusion may occur in the boundary region of the object, the size of the window may not exceed the boundary of the object.

In the following description, it is assumed that the size of the window is set to 3 × 3 for convenience of description. When a window having the center of the object as the center pixel is set for the left image, an average value of pixels in the left image window is obtained using a depth map. Then, the average value is rounded to obtain an integer value. A pixel (corresponding pixel) having a value corresponding to the rounded integer value is found in the object area of the right image, and a window of the same size (3 × 3) having the pixel as the center pixel is set in the right image.

When the window is set in the right image, the normalized cross correlation (hereinafter referred to as NCC) of pixels in the window of the left image and the right image is obtained by moving the window of the right image by one pixel. The normal similarity represents a value obtained by normalizing the degree of similarity of pixels in a window.

For example, an NCC value of a window set with the center of the object of the left image as the center pixel and a window set with the corresponding pixel as the center pixel in the right image is referred to as NCC (k1), and the center of the object of the left image is A NCC value of a window set as a center pixel and a window shifted one pixel to the right from the corresponding pixel of the right image is referred to as NCC (k1-1), and centered on the center of the object of the left image. The NCC value of the window set as the pixel and the window set as the center pixel using the pixel shifted to the left from the corresponding pixel of the right image is referred to as NCC (k1 + 1). Equations 4 and 5 below represent equations for obtaining the NCC.

In Equations 4 and 5, W (x, y) is a brightness value at x, y coordinates in a window, and Wm (x, y) is an average value of luminance of all pixels in the window, W '(x, y). ) Denotes a value obtained by dividing a luminance value in the x and y coordinates by subtracting an average value of luminance for all pixels by a norm value. In Equation 5, the operator 크로스 cross correlates the W _L '(x, y) value of each pixel in the window of the left image and the W _R ' (x, y) value of each pixel in the window of the right image. Represents a cross correlation operation. The d ₀ is a value obtained by rounding a parallax average value of pixels in a window obtained from the window of the left image. k indicates the amount of shift of the center pixel of the window when the NCC value is obtained by shifting the position of the window. In the above example, the case in which the window of the right image is moved using the left image as a reference image is taken as an example. However, when the right image is the reference image according to the implementation example, the window of the left image is moved. You can also implement

Using the three obtained NCC values (NCC (k1-1), NCC (k1), NCC (k1 + 1)), the precise parallax at which the normal similarity is maximized can be estimated. For example, assume that the three NCC values form a secondary parabola.

FIG. 5 is a case where the three NCC values form a secondary parabola of symmetry. If the center pixel difference between the center pixel of the left image window and the right image window is k1, the NCC (k1) value is the value of NCC (k1 + 1) when the center pixel difference is k1 + 1, and the center pixel difference is k1-1. In the case of, the NCC (k1-1) value can be seen in the second parabolic coordinate. In FIG. 5, it can be seen that there is a difference between the parallax k2 at the point where the normal similarity is maximum and the parallax k1 of the left and right window center pixels. This is due to the inter-pixel quantization error in pixel-based stereo vision systems. The parallax k2 at the point where the normal similarity is maximum can be obtained using the secondary parabola (y = ax ² + bx + c) and three NCC values.

FIG. 6 is a diagram schematically illustrating estimating parallax using a high resolution image of a detected object region according to an embodiment of the present invention. 6 shows a depth map image of a left and right original image N × M and a reduced resolution image N / 2 ^l × M / 2 ^l . The lower image represents an image in which a depth map image obtained by locally matching an object region using left and right images and a depth map image of a reduced resolution image are combined.

In order to estimate the parallax more accurately with respect to the detected object, a parallax value k2 obtained using the image having the reduced resolution as described above and a high resolution image of the object area may be used. For example, when using the original resolution image (N × M) for the object region as shown in FIG. 6, the parallax value k2 obtained using the reduced image (N / 2 ^l × M / 2 ^l ) ) Is converted to a corresponding parallax value corresponding to the parallax value k2 in the original image by multiplying the inverse value of the reduction ratio.

In other words, if the reduced resolution to 1/2 multiplied by ^l to 2 ^l value as the equation (6) to the k2, by rounding the value obtained by multiplying an integer value. More accurate parallax may be obtained by locally matching the corresponding parallax value k3 in the original image obtained as described above.

The local matching may use the reference image and window matching as described above. That is, window matching may be used by setting a window for the object region of the left image and a window for the object region of the right image. The candidate interval for finding the corresponding point of the window of the reference image and the relative image for the window matching may vary according to the reliability σ of the subsampled interpolation method performed on the reduced resolution image. For example, a three-fold (99.73%) or four-times (99.9968%) interval of reliability is determined as a candidate interval according to a Gaussian distribution function.

Accordingly, the disparity candidate interval x of the corresponding point for window matching is expressed by Equation 7.

The candidate interval (x) is a value obtained by multiplying the reliability (σ) by the inverse value (2 ^l ) of the reduction ratio from the corresponding parallax value k3 in the original image and rounding the value multiplied by the multiple (N). Has a deviation.

For example, assume that the candidate section x has a section of 'k3 + n' in 'k3-n'. In the case where the left image is the reference image, k3 + n is obtained from a window of the left image having a pixel having k3 as a parallax value as a center pixel and a window of the right image having a pixel having k3-n as a center pixel as a center pixel. The pixels having the parallax value are moved by one pixel until the pixel is the center pixel, and each window is matched to find the parallax having the highest matching cost. Similarly, according to an embodiment, the reference image may be set as the right image.

The window matching function may use Normalized Cross Correlation (NCC) as described above, or may use Sum of Absolute Difference (SAD), Sum of Squared Difference (SSD), or the like. When NCC is used as a matching function, the parallax value that maximizes the NCC value according to window matching can be estimated as the final parallax of the object. When using SAD or SSD as a matching function, SAD or The parallax value that minimizes the SSD value may be estimated as the final parallax of the object. In order to estimate the maximum value or the minimum value, a sub-pixel interpolation method such as a parabola fitting method or an enhanced normalized cross correlation (ENCC) method may be used. In the case of using the parabola fitting, as described above, a parabola can be assumed and the maximum or minimum value can be estimated.

When the final parallax value corresponding to the maximum value or the minimum value is obtained, the distance from the object may be calculated by applying the same to the equations (1) to (3). The calculated distance value is filtered using a dynamic filter such as a Kalman filter or a particle filter to improve accuracy by tracking the distance to the object in units of image frames. Can be.

As described above, the object is detected by global matching of the low resolution image, and the parallax is estimated by locally matching the high resolution image of the detected object, so that accurate parallax can be estimated while reducing the complexity. When a plurality of objects are detected, the parallax estimation process may be performed in series or in parallel on each detected object.

7 is a diagram schematically illustrating a configuration of a distance estimating apparatus according to an embodiment of the present invention. The distance estimating apparatus may include a first image capturing apparatus 700, a second image capturing apparatus 705, a first image processor 710, a second image processor 715, a first matching unit 720, and an object detector. 730, a first interpolator 740, a second matcher 750, a second interpolator 760, a distance estimator 770, and a filter 780.

The first image capturing apparatus 700 and the second image capturing apparatus 705 capture a left image and a right image necessary for estimating distance, speed, etc. in the stereo vision system, respectively. For example, a camera or the like may be used as the image capturing apparatus. The left image and the right image obtained by the first image capturing apparatus 700 and the second image capturing apparatus 705 are transmitted to the first image processor 710 and the second image processor 715, respectively. The first image processor 710 and the second image processor 715 reduce and output the input image at the same ratio.

The left image and the right image output by reducing the resolution of the first image processor 710 and the second image processor 715 respectively are input to the first matching unit 720. The first matching unit 720 may globally match the input left image and the right image to obtain a depth map of the image.

The object detector 730 may extract an object for estimating a distance or a speed using the depth map information. For example, as described above, the object may be extracted using a vertical parallax map method. The first interpolator 740 may estimate the parallax of the object by using a subsampling interpolation method on the object detected by the object detector 730. For example, the first interpolator 740 as described above may perform a parallax estimation process using window matching and a maximum NCC value. The first interpolator 740 may calculate and transmit the disparity candidate interval x information to be used by the second matching unit 750 and the corresponding disparity value k3 in the original image.

The second matching unit 750 locally matches the original image of the detected object part using the corresponding disparity value k3 and the disparity candidate interval x information. The original image is received by the first image capturing apparatus 700 and the second image capturing apparatus 705 and stored in a buffer (not shown). The local matching may use window matching as described above, and may use NCC, SAD, SSD, etc. as a matching algorithm.

The second interpolator 760 estimates the final parallax of the object using a subsampling interpolation method using the local matching result of the second matching unit 750, and the distance estimator 770 estimates the final parallax. Estimate the distance of the object using. The filter 780 may track the distance to the object in units of frames by dynamically filtering the estimated distance of the object. The filter unit 780 may use a dynamic filter such as a Kalman filter or a particle filter.

8 is a flowchart schematically illustrating a distance estimation method according to an embodiment of the present invention.

The image is captured using the left and right image capturing apparatus of the stereo vision system, and the resolution of the captured left and right images is reduced by the same ratio (S800). An object included in the image is detected by global matching using the image having the reduced resolution (S810). For example, a depth map may be obtained by globally matching an image having a reduced resolution, and an object may be detected using a vertical parallax map of the depth map.

When the object is detected, the parallax of the detected object is calculated by performing subsampling interpolation on the detected object area (S820). Since the parallax calculated in step S820 is a parallax for an image having a reduced resolution, a corresponding parallax in the original image is obtained by multiplying the parallax by the inverse of the image reduction ratio.

The original image corresponding to the object region detected in operation S810 is extracted, the object region of the original image is locally matched (S830), and subsampling interpolation is performed using the locally matched result (S840). When the final parallax of the object is calculated through step S840, the distance of the target object is estimated using the parallax (S850). The object distance estimation method using the parallax is as described above. When the distance of the object is estimated in step S850, the distance to the object may be tracked in units of frames by filtering the estimated distance information.

The present invention is not limited to the above-described embodiments, and as can be seen in the appended claims, modifications can be made by those skilled in the art to which the invention pertains, and such modifications are within the scope of the present invention.

1 is a diagram schematically illustrating a distance estimating method according to an exemplary embodiment of the present invention.

2 is a diagram schematically illustrating a process of reducing a resolution of a captured stereo image according to an embodiment of the present invention.

3 is a diagram illustrating a depth map of an image having a reduced resolution as an embodiment according to the present invention;

4A and 4B are diagrams illustrating a process of detecting an object using a depth map, according to an embodiment of the present invention.

5 is a diagram schematically illustrating parallax estimation according to a parabola fitting method according to an embodiment of the present invention.

FIG. 6 is a diagram schematically illustrating estimating parallax using a high resolution image of a detected object region according to an embodiment of the present invention.

7 is a diagram schematically illustrating a configuration of a distance estimating apparatus according to an embodiment of the present invention.

8 is a flowchart schematically illustrating a distance estimation method according to an embodiment of the present invention.

* Explanation of symbols for the main parts of the drawings

700: first video recording device 705: second video recording device

710: first image processor 715: second image processor

720: first matching unit 730: object detection unit

740: First interpolation unit 750: Second matching unit

760: second interpolation unit 770: distance estimation unit

780: filter unit

Claims (20)

Reducing the resolution of the captured stereo image; Detecting an object by using a depth map calculated by globally matching the reduced stereo image; Calculating parallax information by subsampling interpolation on the inner region of the detected object; Matching an area of the detected object by using the parallax information and the original stereo image; Subsampling interpolation and estimating parallax using the result of matching the object region; And Estimating a distance to the detected object using the estimated parallax. The method of claim 1, Reducing the resolution of the stereo image, A distance estimation method of reducing the resolution of the captured left image and the right image by the same ratio. The method of claim 1, Detecting the object, A distance estimation method for detecting an object using a vertical parallax map of a depth map obtained by global matching of a stereo image. The method of claim 1, Computing the time difference information, A method of estimating disparity by window matching and subsampling interpolation of internal pixels of a detected object. The method of claim 4, wherein The window matching, Distance estimation method for obtaining normal similarity value between windows by using Normalized Cross Correlation (NCC) algorithm. The method of claim 1, The time difference information, A method for estimating distance comprising a corresponding parallax value obtained by multiplying an object parallax value by an inverse value of an image reduction ratio in a resolution-reduced image and a disparity candidate interval value according to subsampled interpolation reliability. The method of claim 1, Matching an area of the detected object by using the original stereo image, Window-matching pixels in the object region using one of NCC, Sum of Absolute Difference (SAD) or Sum of Squared Difference (SSD) algorithms. Distance estimation method. The method of claim 1, The subsampling interpolation and estimating the time difference, A method of estimating distance by estimating the value of a point where an NCC value becomes a maximum when an object region is matched using an NCC algorithm. The method of claim 1, The subsampling interpolation and estimating the time difference, A method of estimating distance by estimating the value of a point where the SAD or SSD value becomes the minimum value when the object area is matched using the SAD or SSD algorithm. The method of claim 1, And filtering the distance from the estimated object to track the distance to the object. The method of claim 10, The filtering step, Distance estimation method of tracking the distance from an object in units of frames using a dynamic filter. In the distance estimation device, An image processor which reduces the resolution of the captured stereo image; A first matching unit which globally matches the stereo image with reduced resolution; An object detector for detecting an object by using the depth map calculated by the first matching unit; A first interpolation unit configured to calculate parallax information by performing subsampling interpolation on the inner region of the detected object; A second matching unit which matches the area of the detected object by using the parallax information and the original stereo image; A second interpolation unit which interpolates subsamples and estimates parallax using the result matched by the second matching unit; And And a distance estimator for estimating a distance to the detected object using the parallax estimated by the second interpolator. The method of claim 12, The object detector, A distance estimating apparatus for detecting an object by using a vertical parallax map of the depth map calculated by the first matching unit. The method of claim 12, The first interpolation unit, And a parallax information calculated by window matching and subsampling interpolation of the internal pixels of the detected object. The method of claim 14, The first interpolation unit, A distance estimator for window matching internal pixels using a normal similarity (NCC) algorithm and obtaining a normal similarity value between windows. The method of claim 12, The time difference information, A distance estimating apparatus comprising a corresponding disparity value obtained by multiplying a parallax value of an object by an inverse value of an image reduction ratio in a resolution-reduced image and a disparity candidate interval value according to subsampled interpolation reliability. The method of claim 12, The second matching unit, A distance estimation device for window matching pixels in an object region using one of NCC, SAD or SSD algorithms. The method of claim 12, The second interpolation unit, And the second matching unit estimates the value of the point where the NCC value becomes the maximum value by parallax when the object region is matched using the NCC algorithm. The method of claim 12, The second interpolation unit, And estimating a value of a point at which the SAD or SSD value becomes the minimum value with parallax when the object region is matched using the SAD or SSD algorithm in the second matching unit. The method of claim 12, And a filter unit configured to dynamically track the distance from the estimated object and track the distance from the object in units of frames.

Images