KR101211872B1 - Apparatus and method for realtime eye detection - Google Patents

Abstract

The real-time eye detection apparatus 100 according to an exemplary embodiment of the present invention may be an eye candidate region acquirer 110 or an eye candidate region acquirer 110 that obtains an eye candidate region from an image including a face region. An eye candidate region size converter 120 that generates a plurality of downscaling images for the region, an image of the eye candidate region acquired by the eye candidate region acquirer 110, and an image generated by the eye candidate region size converter 120 are generated. The MCT converter 130 that performs the MCT operation on each image and the MCT converter 130 generate a window for each image on which the MCT operation is performed, and look up the learning data for each pixel constituting the generated window. Using the table, the sum of the reliability values of all the pixels constituting the window of the learning data comparator 140 and the learning data comparator 140 matching the reliability values for each pixel is the final eye of the eye candidate region having a reference value of more than. domain It includes an eye detection unit 150 for detecting a.

Description

Real-time eye detection device and its method {APPARATUS AND METHOD FOR REALTIME EYE DETECTION}

The present invention relates to an apparatus and a method for detecting a human eye in an image. In particular, the present invention relates to an apparatus and method for performing eye detection by generating downscaling images of eye candidate images in parallel.

The general face recognition system refers to a system for finding a face from an input image and analyzing a feature of the face. The face recognition system determines who is a face and recognizes an identity by comparing with a previously stored face database. As a feature point for use in the face recognition system, eye coordinates may be used in the face area, and as the eye coordinates are accurately detected, the recognition performance of the face recognition system may be improved.

  When the general image processing technique is applied, the captured image is displayed in various sizes and shapes. For this reason, the eye detection can be implemented in a relatively simple way under the constraints of the position of the face and the brightness of the name of the face.However, in a general environment, the image is modified according to the change of the size and position of the face and the change of illumination. Make eye detection difficult

  In order to solve this problem, the eye detection method using eye verification and eye position correction (published number 10-2008-0114361) detects eye candidates using eye detection feature point data obtained through AdaBoost learning algorithm in MCT-converted images. In addition, we proposed a method that can perform eye detection robustly to lighting and environmental changes by performing eye candidate verification using pattern correlation based on MCT.

  However, since this method is also performed sequentially based on the software, the face image is scaled to several predetermined sizes, MCT is applied to each scaled image, and the eyes are detected by comparing the stored learning data for each MCT data. Since the execution time becomes very long, the real-time processing is difficult.

The real-time eye detection apparatus and method thereof according to the present invention aims to solve the following problems.

First, the eye position is to be detected quickly in the face region image.

Second, scale images of the right / left eye candidate regions are simultaneously generated, MCT is applied to the generated images in parallel, and Cascade is applied to each MCT applied data in parallel to the eye in the face region of the image in real time. We want to detect the location.

Third, eye detection is promptly performed by acquiring only one right or left eye candidate region.

The solution to the problem of the present invention is not limited to those mentioned above, and other solutions not mentioned can be clearly understood by those skilled in the art from the following description.

An apparatus for detecting real-time eyes according to the present invention includes an eye candidate region obtaining unit obtaining an eye candidate region from an image including a face region and an eye generating a plurality of downscaling images of the eye candidate region obtained by the eye candidate region obtaining unit. MCT operation for performing MCT operation on each image generated by the candidate area size conversion unit, the image of the eye candidate area acquired by the eye candidate area acquisition unit, and the eye candidate area size conversion unit, and MCT operation is performed by the MCT conversion unit Creates a window for each image, and configures the window of the learning data comparison unit and the learning data comparison unit for matching the reliability value for each pixel using the learning data lookup table for each pixel constituting the generated window. And an eye detector configured to detect, as the final eye region, an eye candidate region whose sum of reliability values is equal to or greater than a reference value.

The apparatus for detecting real-time eyes according to the present invention further includes a memory unit in which the input image is stored in units of frames, and the memory unit also stores a plurality of downscaling images generated by the eye candidate region size converter.

When the eye candidate region acquisition unit according to the present invention acquires one of the right or left eye candidate regions, the other eye candidate region is obtained by rotating the acquired image in left and right symmetry.

In the real-time eye detection apparatus according to the present invention, the eye candidate region size converter performs downscaling only for one of the right or left eye candidate regions, the MCT transformer performs the MCT operation, and the confidence value is matched in the training data comparison unit. The eye detection unit detects the final eye region.

The eye candidate region obtaining unit according to the present invention is characterized by simultaneously obtaining image data for the right and left eye candidate regions from the image stored in the memory unit in parallel.

When the eye candidate region size converter according to the present invention acquires the eye candidate image from the eye candidate region acquisition unit, simultaneously generates a plurality of downscaling images having different downscaling ratios for the acquired eye candidate images in parallel. It is done.

The learning data comparison unit according to the present invention is characterized in that matching the reliability value using a plurality of learning data lookup table having a different amount of information with respect to the generated window, using a plurality of learning data lookup table, respectively, It is characterized by generating in parallel.

The eye detector according to the present invention is characterized in that the eye candidate region in which the sum of the reliability values of all the pixels is equal to or greater than the reference value for each of the reliability value sets generated by using the plurality of learning data lookup tables is detected. .

The real-time eye detection method according to the present invention is obtained in steps S1 and S1 in which a plurality of downscaling images are generated for an eye candidate area acquired in step S1 and an eye candidate area acquired in step S1 in an image including a face area. A window for each image in which MCT operation is performed in step S3 and MCT operation is generated for each image generated in step S2 and an eye candidate region image, and for each pixel constituting the generated window Step S5 in which the confidence values of each pixel are matched using the training data lookup table, and in step S5, an eye candidate region whose sum of the reliability values of all the pixels constituting the window of step S4 is greater than or equal to the reference value is determined as the final eye region. It includes.

In the step S1 according to the present invention, when one of the right or left eye candidate areas is acquired, the other one of the eye candidate areas is obtained by rotating the acquired image in left and right symmetry.

Step S1 according to the present invention is characterized in that the image data for the right and left eye candidate regions are simultaneously acquired in parallel in the image including the face region.

In the step S2 according to the present invention, a plurality of downscaling images having different downscaling ratios for the eye candidate images obtained in the step S1 are simultaneously generated in parallel.

In the step S4 according to the present invention, the reliability values are matched using a plurality of learning data lookup tables having different amounts of information with respect to the generated window, and the set of reliability values is parallel using a plurality of learning data lookup tables, respectively. It is characterized in that it is generated by the enemy.

In the step S5 according to the present invention, an eye candidate region having a value obtained by adding reliability values of all pixels to each of a set of reliability values generated using a plurality of learning data lookup tables is determined as a final eye region. .

Real-time eye detection apparatus and method according to the present invention has the following effects.

First, a plurality of downscaling images of eye candidate regions are generated in parallel, and whether or not they are eye regions with respect to the generated images can be detected in real time.

Second, by detecting the eye area in real time, face recognition used in the camera, etc. is quickly performed.

Third, even if only one right or left eye candidate region is correctly detected, the eye position can be detected in the image.

The effects of the present invention are not limited to those mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the following description.

1 is a block diagram schematically illustrating the configuration of a real-time eye detection apparatus according to an embodiment of the present invention.
2 is a diagram illustrating a method of converting the size of an eye candidate region by a real-time eye detection apparatus according to the present invention.
3 is a diagram illustrating a method of performing MCT transformation on an input image by a real-time eye detection apparatus according to the present invention.
4 is a diagram illustrating a method of constructing an eye region window and calculating a reliability value from an MCT-converted downscaling image by a real-time eye detection apparatus according to the present invention.
5 is a diagram illustrating a result of detecting eye coordinates from an input image by a real-time eye detection apparatus according to the present invention.
6 is a flowchart of a real-time eye detection method according to an embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

The terms first, second, A, B, etc. may be used to describe various components, but the components are not limited by the terms, but merely for distinguishing one component from other components. Only used as For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

As used herein, the singular forms "a,""an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It is to be understood that the present invention means that there is a part or a combination thereof, and does not exclude the presence or addition possibility of one or more other features or numbers, step operation components, parts or combinations thereof.

Hereinafter, the real-time eye detection apparatus 100 and its method will be described in detail with reference to the drawings.

Prior to the detailed description of the drawings, it is to be clear that the division of the components in the present specification is only divided by the main function of each component. That is, two or more constituent parts to be described below may be combined into one constituent part, or one constituent part may be divided into two or more functions according to functions that are more subdivided. Each of the components to be described below may additionally perform some or all of the functions of other components in addition to the main functions of the components, and some of the main functions of each of the components are different. Of course, it may be carried out exclusively by. Therefore, the presence or absence of each component described through this specification should be interpreted functionally, and for this reason, the configuration of the components according to the wireless power transmission apparatus of the present invention is within the limits to achieve the object of the present invention. Clearly, it may be different from 1.

1 is a block diagram schematically showing the configuration of a real-time eye detection apparatus 100 according to an embodiment of the present invention.

The real-time eye detection apparatus 100 according to an exemplary embodiment of the present invention may be an eye candidate region acquirer 110 or an eye candidate region acquirer 110 that obtains an eye candidate region from an image including a face region. An eye candidate region size converter 120 that generates a plurality of downscaling images for the region, an image of the eye candidate region acquired by the eye candidate region acquirer 110, and an image generated by the eye candidate region size converter 120 are generated. The MCT converter 130 that performs the MCT operation on each image and the MCT converter 130 generate a window for each image on which the MCT operation is performed, and look up the learning data for each pixel constituting the generated window. Using the table, the sum of the reliability values of all the pixels constituting the window of the learning data comparator 140 and the learning data comparator 140 matching the reliability values for each pixel is the final eye of the eye candidate region having a reference value of more than. domain It includes an eye detection unit 150 for detecting a.

Meanwhile, the real-time eye detection apparatus 100 of the present invention may further include a memory unit 160 in which an input image is stored in units of frames. In addition, for example, a plurality of downscaling images generated by the eye candidate region size converter 120 may be temporarily stored.

The memory unit 160 is an element that acquires an image input from a camera and stores an image of one frame. It is desirable to have a memory capable of random access as a space for storing images, and double buffering is used for eye detection by processing image information stored in the other memory while the image is being written to the memory.

The eye candidate region obtaining unit 110 obtains an eye candidate region from an image input from a camera or an image stored in the memory unit 160. In order to obtain an eye candidate region, an apparatus or method for detecting a face region of a person in an image to be used must be preceded. Based on the fact that there is coordinate information of the face area, an area in which the eye is highly likely to be located is acquired as the eye candidate area.

 The eye candidate region size converting unit is a component that converts a candidate region, which is the object of eye detection, to a predetermined size using face region coordinate information. Since the eye candidate area varies in size depending on the size of the face, it is necessary to convert the image into a fixed size image and then perform the work.

If a 15x15 window is used for eye detection, the detectable eye size is set to 15x15. However, depending on the size of the eye in the face area, the 15x15 window may not contain all of the eye area. To prepare for this case, it is necessary to downscale the image so that all eyes are included in the 15x15 window. That is, the eye candidate region size converter first converts the eye candidate region to a predetermined size and then downscales the converted image at a predetermined ratio.

When downscaling according to various ratios, a resolution difference of each downscale image occurs. Therefore, when face detection is performed using the same clock, a space that does not use a pixel or line clock occurs. The real-time eye detection apparatus 100 according to the present invention may use a blank in which the clock is not used in order to reduce the hardware usage of the face learning data in the learning data comparison unit 140.

2 is a diagram illustrating a method for converting the size of an eye candidate region by the real-time eye detection apparatus 100 according to the present invention.

The eye candidate region obtaining unit 110 receives the stored image and coordinate information of the face region obtained from an external face detection apparatus. In FIG. 2, the eye candidate area is divided into a left eye candidate area and a right eye candidate area. The left eye candidate area becomes the upper left area of the area obtained by dividing the face area into two parts in the top, bottom, left and right directions. The right eye candidate region becomes an upper right region among the divided regions as described above.

In this case, the right eye candidate region may be obtained by rotating the original image region in left and right symmetry. Of course, both eye candidate regions can be obtained, but when one eye candidate region is used, the same MCT transformation and learning data comparison are performed for the left eye region and the right eye region. Save space for storing training data.

The image input in FIG. 2 is stored in a memory of the memory unit 160. Therefore, in order to obtain the left eye and right eye candidate images from the images stored in the memory, the memory must be sequentially accessed. When importing the left eye candidate image, the memory may be accessed from left to right of the face image, and the right eye candidate image may access data from right to left in the face image. That is, the eye candidate region may be obtained in parallel with respect to the right eye candidate image and the left eye candidate image.

 The eye candidate image may be stored in a separate register. Therefore, once the memory access is completed and the left and right eye candidate images are generated, the downscaling images may be simultaneously generated in parallel in the eye candidate region size converter 120. As shown in FIG. 2, when the left eye candidate image and the right eye candidate image are generated, the remaining three downscale images are simultaneously generated in parallel.

As a result, MCT (Modified Census Transform) transformation is performed on the eye candidate region acquired by the eye candidate region acquisition unit 110 and the downscaling image generated by the eye candidate region size transformation unit 120. In FIG. 2, eight eye candidate images are input to the MCT converter 130 at the same time to generate eight MCT images.

When the eye candidate region size converter 120 acquires the eye candidate image by the eye candidate region acquirer 110, simultaneously generates a plurality of downscaling images having different downscaling ratios for the acquired eye candidate images in parallel. do.

The MCT converter 130 performs an MCT operation on the eye candidate original image acquired by the eye candidate region acquirer 110 and the downscaled image by the eye candidate region size converter 120. The MCT operation generates a 3x3 window and compares the average of all nine pixel values with each pixel brightness value and calculates it as 0 if it is less than the average and 1 if it is greater than or equal to the average. The calculation is performed on a total of nine pixels, resulting in a 9-bit long code.

3 is a diagram illustrating a method of performing MCT transformation on an input image by the real-time eye detection apparatus 100 according to the present invention.

3 is a diagram illustrating a method of performing MCT transformation on an input image by the real-time eye detection apparatus 100 according to the present invention. The average value of the pixels in the 3x3 window shown in FIG. 3 is 95.11. The average value is compared with each pixel in the window and converted to 0 when the pixel value is small and 1 when the pixel value is large or the same. The concatenated bits are then concatenated in order to generate a 9-bit bit string. At this time, the obtained 9-bit bit string becomes the MCT bit string value of the center pixel in the 3x3 window. In this way, the MCT converter 130 performs MCT conversion using the brightness values of pixels in the 3x3 window.

In this case, the combination order of the generated bits should be the same as the order used in the process of generating the learning data used in the learning data comparison unit 140.

The training data comparison unit 140 determines the reliability of each pixel constituting the window by using each window image and previously prepared training data. Learning data is organized in the form of lookup tables.

The training data lookup table is a lookup table previously learned based on the brightness of each pixel constituting the face region image. The face area around the eyes differs in brightness depending on the height and curvature of the parts that make up the face.The learning data lookup table is a face area according to brightness for each pixel using an image that is input in advance for learning. Is stored as a probability value. The confidence value means this probability value.

For example, if the brightness of a specific pixel is learned to be 23 in the face image, when the brightness of the corresponding pixel is 23, the confidence value is 1, and when the brightness is 15, the confidence value is about 0.68, and the brightness is about 15. Is 150, a reliability value of about 0.15.

The training data comparison unit 140 receives an MCT image of the downscaled eye image and generates a window having a predetermined size. The window having a predetermined size sequentially moves along the input MCT image. The learning data comparison unit 140 compares the image included in the window with the already generated learning data and determines whether the image included in the window is an eye region.

  In the learning data comparison unit 140, windows are implemented based on registers, and the windows of each window may be simultaneously accessed. This structure makes it possible to obtain confidence values for all pixels in a window at the same time. The pixels at each position in the window maintain a confidence value according to the MCT value of the corresponding pixel in the form of a lookup table.

4 is a diagram illustrating a method of configuring an eye region window and calculating a reliability value from an MCT-converted downscaling image by the real-time eye detection apparatus 100 according to the present invention.

4 is a diagram illustrating a method of constructing an eye region window and calculating a reliability value from an MCT-converted downscaling image by the apparatus 100 for real-time eye detection according to the present invention. 4 shows only an example for the left eye and the right eye is performed simultaneously with the left eye detection in the same manner.

On the other hand, the learning data comparison unit 140 may be composed of several stages. For example, in stage 1, when the 22 pixels are selected in a 15x15 window, the reliability value of each pixel is extracted through the lookup table, and when the reliability value obtained by adding all the reliability values is higher than or equal to a predetermined reference value, the input image may be an eye image. Judgment is made to proceed to the next stage (for example, stage 2). After that, the sum of the reliability values acquired in each stage becomes the final reliability value of the corresponding image area in the case of the image passing through all the stages.

  The eye candidate region size converter 120 according to the present invention generates a plurality of downscaling images according to the method shown in FIG. 2. These downscaling images are transformed into images consisting of MCT bit streams through the MCT converter 130. From the first downscaling image to the fourth downscaling image illustrated in FIG. 4, the image is converted into a mode MCT bit string. As shown in FIG. 4, all downscaling images are input in parallel from stage 1 to stage 3. Each stage may determine whether an area in which a 15x15 window is located is an eye area for all downscaling images during one input clock period (or a predetermined clock period).

Eventually, the learning data comparison unit 140 matches the reliability values using a plurality of learning data lookup tables having different amounts of information with respect to the generated window, and sets the reliability values using each of the plurality of learning data lookup tables. Can be generated in parallel.

The eye detector 150 uses the confidence value obtained by the learning data comparator 140 when the confidence value is equal to or greater than a specific value, considering the position in the MCT image of the window and the position in the entire image of the MCT image. The eye area is detected in the image. In particular, while moving the MCT image in which the window is input, the same eye region may be overlapped and detected. As such, it is preferable that the overlapped eye region is selected as the final eye region having the largest confidence value according to the training data.

The eye detector 150 detects, as the final eye region, an eye candidate region having a sum of reliability values of all pixels equal to or greater than a reference value for each of the reliability value sets generated using the plurality of learning data lookup tables. The confidence value set refers to confidence value data for each pixel generated using one learning data lookup table.

5 is a diagram illustrating a result of detecting eye coordinates from an input image by the real-time eye detection apparatus 100 according to the present invention.

In stage 1 of FIG. 5, the window moves from left to right and top to bottom along the MCT transformed image. As the window moves, it brings the reliability of the MCT value for a particular pixel within the window. Reliability data is stored in advance in the lookup table. If the sum of the reliability of the pixels in the window becomes the reliability at the current window position, and the images that do not meet the reference condition such that the reliability is above (or below) the reference value are judged not to be the face region, and further progress the stage. I never do that.

  Stage 2 evaluates the reliability of the downscaled image by using a lookup table with more information than Stage 1. Like Stage 1, Stage 2 also calculates a reliability value to determine whether the value advances to the next stage based on the reference value or the reference condition. The reference value may be determined in advance using the learning data, or may be set according to a system or user setting.

  After passing through all stages, the window area is determined to be an eye area. The real-time eye recognition apparatus according to the present invention does not perform the individual stages sequentially, but simultaneously performs an operation and determines an area that passes all three stages as an eye region.

  Each stage's window and lookup table consist of separate registers. Instead of a sequential approach to all the pixels in the window, the confidence value can be obtained from the lookup table. Therefore, compared to the software method of the conventional PC, the number of memory accesses can be reduced, thereby reducing the time for calculating the total reliability value.

5 is a diagram illustrating a result of detecting eye coordinates from an input image by the real-time eye detection apparatus 100 according to the present invention. In FIG. 5, a total of two faces are detected to detect left and right eyes for each face. There may be a large number of areas judged as eyes among the images photographed with the 15x15 window size. For example, all windows including eye ⓐ in face ① are likely to be detected as eyes. That is, several eyes can be detected near eye ⓐ. In this case, the eye detection unit 150 according to the present invention selects a window having the largest reliability sum of the eye region among the eye candidate groups detected near eye ⓐ as the final eye region. In the detection result of FIG. 5, the center point was output as a red dot in the detected eye region.

Hereinafter, a real-time eye detection method according to the present invention will be described. Details overlapping with the above-described real-time eye detection apparatus 100 will be briefly described. 6 is a flowchart of a real-time eye detection method according to an embodiment of the present invention.

The real-time eye detection method according to the present invention is obtained in steps S1 and S1 in which a plurality of downscaling images are generated for an eye candidate area acquired in step S1 and an eye candidate area acquired in step S1 in an image including a face area. A window for each image in which MCT operation is performed in step S3 and MCT operation is generated for each image generated in step S2 and an eye candidate region image, and for each pixel constituting the generated window Step S5 in which the confidence values of each pixel are matched using the training data lookup table, and in step S5, an eye candidate region whose sum of the reliability values of all the pixels constituting the window of step S4 is greater than or equal to the reference value is determined as the final eye region. It includes.

In operation S1, when one of the right or left eye candidate regions is acquired, the other eye candidate region may be obtained by rotating the acquired image in left and right symmetry.

Of course, both eye candidate regions may be acquired respectively, and in this case, step S1 may simultaneously acquire image data for the right and left eye candidate regions in the image including the face region in parallel.

In step S2, a plurality of downscaling images having different downscaling ratios for the eye candidate images acquired in step S1 are simultaneously generated in parallel.

In step S4, the reliability values are matched using a plurality of learning data lookup tables having different amounts of information with respect to the generated window, but a set of reliability values is generated in parallel using each of the plurality of learning data lookup tables. .

In step S5, the eye candidate region in which a reliability value of all pixels is added to each of the reliability value sets generated by using the plurality of learning data lookup tables is determined as the final eye region.

The embodiments and drawings attached to this specification are merely to clearly show some of the technical ideas included in the present invention, and those skilled in the art can easily infer within the scope of the technical ideas included in the specification and drawings of the present invention. Modifications that can be made and specific embodiments will be apparent that all fall within the scope of the present invention.

100: real-time eye detection device 110: eye candidate region acquisition unit
120: eye candidate region size conversion unit 130: MCT conversion unit
140: learning data comparison unit 150: eye detection unit
160: memory unit

Claims (14)

In the device for detecting the eye in the image,
An eye candidate region obtaining unit obtaining an eye candidate region in an image including a face region;
An eye candidate region size converter configured to generate a plurality of downscaling images of the eye candidate region acquired by the eye candidate region obtaining unit;
An MCT conversion unit performing an MCT operation on an image of an eye candidate region obtained by the eye candidate region obtaining unit and each image generated by the eye candidate region size converting unit;
The MCT converter generates a window for each image on which the MCT operation has been performed, and compares learning data that matches a reliability value for each pixel using a learning data lookup table for each pixel constituting the generated window. part; And
And an eye detector configured to detect an eye candidate region having a sum of reliability values of all pixels constituting the window of the training data comparison unit as a final eye region as a final eye region. The method of claim 1,
And a memory unit configured to store the input image in units of frames, wherein the memory unit also stores a plurality of downscaling images generated by the eye candidate region size converter. The method of claim 1,
The eye candidate region obtaining unit
And when one of the right or left eye candidate areas is acquired, the other eye candidate area is obtained by rotating the acquired image left and right symmetrically. The method of claim 3,
Only for one of the right or left eye candidate regions
The eye candidate region size converter performs downscaling, the MCT transformer performs an MCT operation, matches the confidence value in the training data comparison unit, and detects the final eye region in the eye detector. Eye detection device. The method of claim 2,
The eye candidate region obtaining unit
And image data for the right and left eye candidate regions are simultaneously obtained in parallel from the image stored in the memory unit. The method of claim 1,
The eye candidate region size converter
And when the eye candidate region acquisition unit acquires the eye candidate image, simultaneously generating a plurality of downscaling images having different downscaling ratios for the acquired eye candidate images in parallel. 7. The method according to claim 1 or 6,
The learning data comparison unit
Characteristic values are matched using a plurality of learning data lookup tables having different amounts of information with respect to the generated window, and a set of reliability values is generated in parallel using a plurality of learning data lookup tables, respectively. Real-time eye detection device. The method of claim 7, wherein
The eye detection unit
And an eye candidate region in which a sum of reliability values of all pixels is equal to or greater than a reference value for each of the reliability value sets generated by using the plurality of learning data lookup tables is detected as the final eye region. In the method for detecting the eye in the image,
Step S1 of acquiring an eye candidate region from an image including a face region;
Step S2 of generating a plurality of downscaling images with respect to the eye candidate area acquired in step S1;
An SCT step of performing an MCT operation on the eye candidate region image acquired in the step S1 and each image generated in the step S2;
Step S4 of generating a window for each image on which the MCT operation is performed in step S3, and matching a reliability value with respect to each pixel using a learning data lookup table for each pixel constituting the generated window; And
And a step S5 in which an eye candidate region having a sum of reliability values of all pixels constituting the window of step S4 equal to or greater than a reference value is determined as the final eye region. 10. The method of claim 9,
The step S1
And when one of the right or left eye candidate areas is acquired, the other eye candidate area is obtained by rotating the acquired image in left and right symmetry. 10. The method of claim 9,
The step S1
Real-time eye detection method, characterized in that to acquire the image data for the right and left eye candidate region in parallel in the image including the face region. 10. The method of claim 9,
The step S2
And a plurality of downscaling images having different downscaling ratios for the eye candidate images acquired in step S1 are generated in parallel at the same time. The method according to claim 9 or 12,
The step S4
Reliability values are matched using a plurality of learning data lookup tables having different amounts of information with respect to the generated window. A reliability value set is generated in parallel using a plurality of learning data lookup tables, respectively. Real time eye detection method. The method of claim 13,
The step S5
And an eye candidate region in which a reliability value of all pixels is added to each of a set of reliability values generated using the plurality of learning data lookup tables is determined as a final eye region.

Images