KR20080101388A - NMC-based face detection method - Google Patents

Abstract

The present invention relates to a NMCT-based face detection method, and (a) to detect faces of various sizes in the pre-processed image to extract the intensity information by converting the input color image to a black and white image Resizing the image to a maximum and minimum reduction ratio to adjust a detectable maximum and minimum face size; (b) a new MCT transformation step of transforming the resized image using the local structural features of the image; (c) a candidate detection step of detecting a face by applying a face detection window by importing coordinates of each stage filter table to the image converted into the regional structural features; (d) a grouping step of outputting a final result by grouping the detected face areas of the detected candidates of various sizes and positions by the same face candidate area, and (e) accumulating classification of face candidate areas that are representative of each group. The method includes a candidate registration step of applying a child value as a weight to obtain a weighted average of a representative face candidate region and a center pixel coordinate, and selecting a face candidate region as a result. The present invention provides a device that enables a user to quickly detect, verify, and recognize a user's face, and provides a toy using the same.

Description

NM-based face detection method {A Face Detection Algorithm based on a New Modified Census Transform}

1 is an embodiment according to the present invention, a hardware configuration for detecting and verifying a face based on a New Modified Census Transform (MCT), and a block implemented in the form of a System on Chip (SoC). It is also.

2 is a block diagram showing the detailed configuration of the NMCT-based face detection apparatus of the present invention.

3 is a diagram illustrating an example of grouping with the algorithm of the present invention.

4 is a flowchart illustrating an algorithm for face detection and verification in the present invention.

5 illustrates a state in which a toy is manufactured by applying an NMCT-based face detection apparatus according to the present invention.

♣ Explanation of symbols for main part of drawing ♣

10: CPU 11: first storage unit

12: second storage unit 13: camera

14: Monitor 15: RS232 Interface

16, 66: JTAG interface 17, 67: expansion connector

18: audio input / output unit 19: button input unit

20: Save face button 22: Face information reset button

23: Charging adapter jack 25: Convex mirror

30: bridge 50: interface

60: first memory unit 61: first memory unit controller

62: second memory unit 70: controller

71: preprocessing module 72: resize module

73: NMCT module 74: candidate detection module

75: grouping module 76: filter data storage unit

100: main board 200: FPGA board

The present invention relates to a NMCT-based face detection method, and more particularly, to NMCT-based face detection and recognition of a user's face within a short time by using memory efficiently to be suitable for hardware implementation. Relates to a face detection method.

In general, various researches have been conducted on biometric technologies such as fingerprint recognition, iris recognition, and vein recognition. Among such biometric technologies, face detection technology is in the spotlight because it is less coercive to a user than other biometric information and has a low contact resistance. In particular, face detection has been recognized and studied as a previous step technique for face verification or face recognition, but is actively being actively researched due to its importance and various commercial applications.

The face area detection methods studied so far can be classified into four types. That is, template matching based on a specific template form, skin color based method using skin color, neural network method mimicking the way human face is perceived, and feature based approach using structural features.

First, template matching is a method of detecting a face by making some standard templates and comparing them with an input image. This method has a disadvantage in that it is difficult to generate a general template for various facial shapes.

The skin color based method is a method of detecting from an input image using a unique skin color of a person. This method can be used quickly and easily, but there is a limitation that the face detection is weak when there is a color change by lighting or a color similar to the skin color in the background. In addition, the neural network method is a way of thinking about how a person perceives another person's face and modeled it. It is a method of distinguishing a face from a non-face pattern through learning on a multi-layer neural network. However, it is possible to detect the learned face more effectively than other methods, but it is difficult to detect various face changes such as rotation and it is not suitable for real-time face detection method due to the large amount of calculation. In addition, the feature-based approach is a method for detecting a face by using unique features of the face such as local features and structural features of the face. This method has a disadvantage in that it has a large amount of calculation like neural network method, but it has an advantage in that face detection is possible by varying various elements according to the characteristics used.

The present invention is a face region detection method, which uses an algorithm based on a feature-based approach, that is, a New Modified Census Transform (MCT) transform that transforms a CT (Census Transform) transform, which is a transform method using a local structural feature of a face. An object of the present invention is to provide a face detection method based on NMCT, which detects a face by efficiently converting memory to be suitable for hardware implementation.

Furthermore, the present invention makes it less susceptible to light changes such as lighting, which is a major problem when detecting a face using a new MCT transformation method, and is an efficient resource for hardware implementation using a filter trained with a front face image. Another object is to provide a face detection method based on NMCT.

Another object of the present invention is to provide an intelligent smart robot toy in which the new MCT transformation method is applied to a toy doll.

The present invention provides a face detection method based on NMCT, which comprises (a) converting an input color image into a black and white image to extract faces of various sizes from a preprocessed image of extracting intensity information only. Resizing the image to a maximum and minimum reduction ratio to adjust the maximum and minimum face sizes to be detected; (b) a new MCT transformation step of transforming the resized image using the local structural features of the image; (c) a candidate detection step of detecting a face by applying a face detection window by importing coordinates of each stage filter table to the image converted into the regional structural features; (d) a grouping step of outputting a final result by grouping the detected face areas of the detected candidates of various sizes and positions by the same face candidate area, and (e) accumulating classification of face candidate areas that are representative of each group. And a candidate registration step of selecting a face candidate region as a result by obtaining a weighted average of the size of the representative face candidate region and the center pixel coordinates by applying the child value as a weight.

In addition, the present invention provides a CPU for controlling data input and output from the components provided on the motherboard, a first storage unit for inputting and outputting data required for various files and operations, and a system program including an OS for booting the motherboard. And a second storage unit for inputting / outputting a camera, a camera for capturing an external image and converting the image into image data, and a display device for visually representing the image data captured by the camera. A controller including a main board including a controller connected to the main board to control data input / output of components mounted on the FPGA board, a first memory unit for inputting / outputting filter data and image data stored under control of the controller; And a second method for storing intermediate results for face detection, calculation, and verification under control of the controller. Including the FPGA board having Li added characterized in that configured.

Hereinafter, with reference to the accompanying drawings, a preferred embodiment of the present invention, NMCT-based face detection method will be described in detail as follows.

1 is a hardware configuration for detecting and verifying a new MCT-based face according to the present invention, and a block diagram showing a main configuration of the main board 100 and the FPGA board 200 connected to the main board and a connector. to be. The block diagram of FIG. 1 implements a new MCT-based face detection method algorithmized by software in the form of a System on Chip (SoC) in a hardware configuration.

First, the motherboard 100 includes a CPU 10, a first storage unit 11 and a second storage unit 12, one or more cameras 13, a monitor 14, an audio input / output unit 18, and a button input unit. 19 is configured, and RS232 interface 15, JTAG interface 16, and expansion connector 17 are provided for connection between main board 100 and FPGA 200.

The CPU 10 controls data input / output and the like from the elements configured on the main board 100. The first storage unit 11 inputs and outputs data necessary for various files and operations, and the second storage unit 12 receives the main data. Input and output of a system program including an OS for booting the board 100. The camera 13 captures an external image, converts the image into image data, and outputs the image data. The monitor 14 displays an input image and a result image as a display device for visually representing image data captured by the camera. The recognition of the face may be improved by configuring the camera 13 in plurality. The monitor 14 may be LCD, organic EL or CRT. In addition, the audio input / output unit 18 converts the operation state of the main board 100 into an audible signal including voice or outputs the converted audio signal from the outside, and includes a microphone and a speaker. The button input unit 19 may include a button necessary to turn on / off power applied to the main board 100, to store and delete a recognized face, or to drive the main board 100.

The component mounted on the field programmable gate array (FPGA) board 200 includes a controller 70 that controls data input / output of the components mounted on the FPGA 200, and includes a first data storing filter data and image data. A memory unit 60 and a second memory unit 62 for storing intermediate results for face detection, calculation and verification are included. In addition, a JTAG interface 66 and an expansion connector 67 are provided as a connection means for data input / output with the main board 100.

2 is a block diagram showing the detailed configuration of the NMCT-based face detection apparatus of the present invention.

The FPGA 200 includes a bridge 30 for synchronizing data transmission and reception with the CPU 10 included in the main board 100, and decodes data input and output to the bridge 30. An interface 50 including a configuration registration and decoding of an address, a decoding and registration decoding of an on-chip memory address, and a file is provided.

The controller 70 mounted on the FPGA 200 is for controlling components mounted on the FPGA 200 for NMCT-based face detection, calculation, and verification. The first memory controller 61 controls input / output of data of the first memory unit 60 in which filter data and image data are stored. The pre-processing module 71 is responsible for converting color image information into a black and white image, and the resize module 72 gradually reduces the image to detect faces of various sizes. It has a function to make it. The NMCT module 73 converts the reduced image into a form for applying the face detection filter, and the candidate candidate module 74 applies the face detection filter to the converted image to detect the face. The detection function is included, and the grouping module 75 is responsible for grouping the detected face areas by various sizes and positions and outputting a final result. The filter data storage unit 76 temporarily stores the filter data input from the first memory unit 60.

When the resize module 72 including the preprocessing module 71, the NMCT module 73, the candidate detection module 74, and the grouping module 75 perform face detection, calculation, and verification, respectively, the second memory unit ( 62) data is stored and output. The second memory unit 6 stores the data reduced in the resize module 72, stores the data converted in the NMCT module 73, stores the candidate detected in the candidate detection module 74, and the grouping module. Data 75, results, and operation information for the indexes of the grouped face regions and equivalent groups are stored in (75), and the filter information, the filter value, and the filter XY value are stored.

Next, the NMCT-based face detection algorithm of the present invention will be described in detail.

는

을 제외한 3×3 픽셀 크기의 지역적 구조내의 이웃 픽셀이고,

는 픽셀의 강도 값이다. 또,

는 비교 함수이며,

일 경우에 1, 그 외의 경우에는 0을 나타낸다.

는 연결(concatenate) 연산자이고, 이것을 이용하여 픽셀

에서의 CT 변환을 정의하면 다음의 수학식과 같다.[CT transform of image] A CT (census transform) transformation is a method of transforming an image by comparing pixel intensity values between a center point of a region and neighboring points. The CT transform is not limited in size, but is limited to a 3x3 pixel size area for comparison with the proposed algorithm.

Is

Neighboring pixels within a 3 by 3 pixel local structure excluding

Is the intensity value of the pixel. In addition,

Is a comparison function,

1 in the case of 1 and 0 in the other cases.

Is the concatenate operator, which uses pixels

The CT transform at is defined by the following equation.

의 모든 비트는 같은 중요도를 갖고, 영역의 중심 값은 0이다. 즉 3×3 픽셀 크기의 지역적 구조의 픽셀 값들이 8-비트의 지역적 구조 특징으로 변환된다.Equation

All bits of are of equal importance, and the center of the region is zero. In other words, pixel values of a 3 × 3 pixel local structure are converted into 8-bit local structure features.

MCT Transformation of Images MCT (modified census transform) transformations are a few modifications to overcome the drawbacks of CT transformation. In other words, the CT transform compares the intensity value of the center pixel with the intensity value of neighboring pixels, and converts the center pixel's conversion value to 0. Therefore, it cannot express all the features of the 3 × 3 pixel size region. There is this. The MCT transformation overcomes this by using an average intensity value in a 3 × 3 pixel size region.

는

을 포함한 3×3 픽셀 크기의 지역적 구조 내의 이웃 픽셀이고,

는

를 포함한 지역적 구조 내의 평균 강도 값이다. 이를 이용하여, 수학식 1을 수정한 MCT 변환을 정의하면 다음의 수학식과 같다.

Is

Neighboring pixels within a 3 by 3 pixel local structure containing

Is

Average intensity value in the local structure, including Using this, defining an MCT transformation in which Equation 1 is modified is as follows.

That is, pixel values of the 3x3 pixel sized local structure are converted into 9-bit local structure features.

[New modified census transform (NCTCT) transform] The existing MCT transform requires 9-bits for each pixel of the image since the points in the 3x3 pixel size region are converted to 0 or 1, respectively. Since the storage of hardware is divided into 8-bit units, it is not suitable for hardware implementation. Therefore, in the algorithm of the present invention, the number of bits required for the conversion to suit the hardware implementation is adjusted as follows.

는

을 제외한

와 맞닿아 있는 이웃 픽셀이고,

가

일 경우에 0,

일 경우에 2, 그 외의 경우에는 1을 나타내는 함수라 하면, 이를 이용하여 NMCT 변환을 정의하면 다음의 수학식과 같다.

Is

excluding

Neighboring pixels that are in contact with

end

0, if

In one case, a function representing 2 and in other cases, a NMCT transform is defined using the following equation.

That is, the pixel values of the 3x3 pixel sized local structure are converted into 7-bit local structure features.

The algorithm proposed in the present invention uses a local structure feature of a 3 × 3 pixel size region in an image. In the case of the CT transform and the MCT transform, the feature was made using all points in the 3x3 pixel size region, but the algorithm proposed in the present invention was made using only five points in the 3x3 pixel size region. That is, in the case of CT transform, all pixels except the center pixel in the 3 × 3 pixel size area have a value of 0 or 1, so 2 ⁸ = 256 features are obtained. Using all the pixels, 2 ⁹ -1 = 511 features are available. The reason why one is excluded is that the converted values of the pixels in the 3x3 pixel size region do not all have one.

In the new MCT transformation proposed in the present invention, since four points have values of 0, 1, and 2, 3 ⁴ = 81 features are obtained. The algorithm provided by the present invention achieves similar performance using fewer features.

The following is a sample generation for face detection according to the present invention. A sample for filter learning is created using a face image and a faceless image of the front face which are randomly extracted. First, the image is transformed according to the size of the face detection window for which an arbitrary front face image is to be created. The size of the face detection window uses a 22 × 22 pixel size as in the conventional method. The deformed image is normalized using the position of the left eye and the distance between both eyes. A new MCT transform is applied to this normalized image to register as a true sample. The image without any face is cropped to fit the size of the face detection window used previously. The image is registered as a false sample by applying a new MCT transform.

The generated sample was trained to filter for face detection. The learning method used the same method as the conventional method. As a result of the training, a filter table is generated, which contains the coordinates of the significant pixels for determining the face area for a 22 × 22 pixel face detection window, with each pixel having 81 classifier values. In order to improve the detection speed, all pixel coordinates are divided into a total of six levels, and there are overlapping coordinates for each step. Each step contains 8, 16, 32, 64, 128, and 269 coordinates.

The flow of the algorithm for face detection of the present invention can be represented by the flowchart of FIG. 4. The algorithm proposed in the present invention is divided into a process of detecting a face from an image (A) and a process of verifying a detected face (B), and simultaneously perform a face detection process and a face verification process.

The overall algorithm flow is to preprocess the input image first, and then gradually reduce the image to detect faces of various sizes in the preprocessed image. After performing a new MCT transformation on the reduced image to convert it into a shape for face detection, a face candidate filter is detected using a face detection filter. Finally, the face candidate regions detected through grouping are grouped, and a representative face candidate for each group is created and registered as a result.

Face detection process (A) is the preprocessing step (S100), the algorithm proposed in the present invention uses the intensity information of the pixel. However, since a color image including R (red), G (green), and B (blue) must be detected, a process of extracting only intensity information by converting the image into a black and white image is necessary.

Then, as a resizing step (S101) for face detection, the image is reduced in a constant ratio so as to detect faces of various sizes in the preprocessed image. When the face detection window of 22 × 22 pixel size is restored to the original image in the reduced image, it becomes the minimum and maximum detectable face size according to the reduced degree. This can be used to control the precision of detection by changing the reduction ratio, and to adjust the maximum and minimum face sizes that can be detected by changing the maximum and minimum reduction ratios.

In addition, as a new MCT transformation step (S102), a new MCT transformation using local structural features of the image is applied to the reduced image and transformed. The face is detected by applying a face detection window to the converted image. The detection method is as the candidate detection step (S103) as follows. After reading the coordinates from the filter table for each step, the conversion value of the corresponding pixel in the face detection window is read. The classifier values in the filter table corresponding to these values are read and accumulated. If all points in the filter table are accumulated step by step, the comparison is made with a predefined face detection threshold value. As a result of the comparison, if it is determined that the face area is determined, the process proceeds to the next step. If it is determined that it is not the face area, the face detection window is moved by one pixel and then applied again.

Next, in the grouping step (S104), the candidates detected in the step (S103) are duplicated detections because the results are found in various sizes and positions by the same face. Therefore, after grouping the same face candidate areas, a representative face candidate is created for each group, resulting in a result.

The grouping step S104 is performed using the distance between the center pixels of the candidate areas detected in the previous step and the size of the candidate area. Comparing each other from the first detected face candidate area to the last detected face candidate area, the face candidate areas having a distance between the center pixel coordinates closer to the threshold and having similar sizes are grouped into the same group. If the number of candidate areas in the group is less than or equal to a predetermined number, it is determined as a background area, and the group is removed. This is shown in FIG. 3.

Finally, a candidate registration step (S105) of selecting a face candidate area representing each group is performed by applying a cumulative classifier value of the face candidate areas of each group as a weight. Using the accumulated classifier value, the weighted average of the size of the representative face candidate region and the center pixel coordinates is obtained and registered as a result.

Since the filter table used in the present invention was made using the front image, it was shown that the detection was somewhat weak in the case of face distortion such as tilting. In addition, detection was vulnerable if the face was cut off at the end of the image, if it was too small or large, or if the face was obscured, such as large sunglasses, or if the specular features of the face were not reflected due to severe light reflections on the glasses. . Except for this, the front image showed very good detection performance.

The most important factor when implementing an image processing algorithm in hardware is the memory size. Since video data requires a lot of memory, using it efficiently is an essential consideration when implementing hardware. 9-bits are required to represent 511 classifiers, and 7-bits are required to represent 81 classifiers. Therefore, 9-bit is a waste of resources due to the nature of the hardware memory stored in 8-bit units. In addition, the existing algorithm requires about 483K of storage space at 484 × 511 × 2 = 494648, and the proposed algorithm requires about 517 × 81 = 41877 and about 41K storage space. do.

In addition, the face detection process is a repetition of a simple process of reading the pixel values corresponding to the coordinates of each step of the face detection window, reading the classifiers from the filter table, accumulating them, and comparing them with the threshold. This makes it easy to implement in hardware and is well suited for the nature of the hardware that handles regular and repetitive tasks quickly and efficiently. Therefore, the algorithm proposed in the present invention uses less memory resources compared to the conventional method, and it is not only efficient to use memory resources when implemented in hardware, but also easy to implement and shows excellent performance.

Furthermore, at the same time as the face detection process, the detected face is verified. This is a face verification process (B) in which a face registered as a candidate calculates a face region through a step S201 of calculating coordinates for a face region, that is, an eye, a nose, and a mouth. In step S202, the facial region is detected. In detecting the face region (S202), a module applied to face detection is used. And comparing and combining the distance of the eyes, nose and mouth of the face area (S203) is performed, and converting the image such as reduction, enlargement and angle of the face area (S204) through all the applied to the face detection The module uses to perform the verification operation (S205). The candidates are registered and registered through the accumulation operation step S206.

As described above, the toy as shown in FIG. 5 may be manufactured by applying a device to which the NMCT-based face detection algorithm is applied.

The toy 300 of FIG. 5 includes one or more cameras 13 for capturing a user and converting the image data into image data, and one or more convex lenses 25 are provided on one side of the camera 13. The convex lens 25 is for enabling the user to adjust the position of the face of the user so that the user can come to the center of the camera 13. Instead of the convex lens 25, one or more concave lenses may be applied. A monitor 14 is provided to display a photographed user's face, and a power supply unit for supplying power for the operation of the toy 300 is provided, and the power supply unit is coupled by inserting a battery or a rechargeable battery or externally. It may be provided with a charging adapter jack 23 or a connector to apply a DC power from.

And the operation of the toy 300, for example, the face storage button 20 for storing the recognized user's face is provided, the monitor 14 or one side of the monitor 14 to delete the first used or stored face information Face information reset button 22 and the like are included. In addition, the toy 300 may include a power button for turning on or off the toy. In addition, a microphone, a speaker, or the like is mounted as the audio input / output unit 18 for outputting voice from the toy 300 or inputting and storing external voice.

The toy 300 configured as described above presses the face storage button 20 in order to store the user's face. For example, the voice “I'm ○○○, let's be friends ~” is output, and the face storage button is stored for the face. Press 20 to register and store face information. In addition, if the user does not store enough face information and releases the face save button, for example, a voice of “no hate ~ play more ~” may be output to induce additional face information storage.

In addition, when a face not registered in the air for face recognition is detected, for example, a voice is outputted, "Who are you? When a face already registered is detected, for example, a voice is outputted, "My friend ~ What should I play with?"

The configuration of the toy 300 is an embodiment of the device applying the NMCT-based face detection algorithm of the present invention, it can be produced and applied in various forms or shapes for various information security, authentication, monitoring or confirmation, etc. will be.

While the invention has been shown and described in connection with specific embodiments thereof, it is well known in the art that various modifications and changes can be made without departing from the spirit and scope of the invention as indicated by the claims. Anyone who owns it can easily find out.

As described above, the present invention has proposed a new MCT-based face detection algorithm that is resistant to changes in illumination for hardware implementation, reduces the waste of memory by adjusting the number of storage bits to be suitable for hardware implementation, and detects small face. By creating and using a filter table, the memory usage is reduced by about 1/12 compared to the conventional method. Although the present invention uses less memory, good detection performance has been obtained.

In addition, the face is detected by scanning more pixels, but by dividing into several stages, the number of candidates dropped in the middle stage is increased, and the speed is improved. There is an effect of detecting the area.

Claims (9)

(a) The maximum and minimum face size that can be detected by converting the image to the maximum and minimum reduction ratio so that the face of various sizes can be detected in the preprocessed image that converts the input color image into a black and white image to extract only intensity information. Resizing step of adjusting; (b) a new MCT transformation step of transforming the resized image using the local structural features of the image; (c) a candidate detection step of detecting a face by applying a face detection window by importing coordinates of each stage filter table to the image converted into the regional structural features; (d) a grouping step of outputting a final result by grouping the detected face regions by various sizes and positions by the same face candidate regions; (e) a candidate registration step of selecting a face candidate region as a result by obtaining a weighted average of the size of the representative face candidate region and the center pixel coordinates by applying the accumulated classifier values of the face candidate regions representing the respective groups as weights; NMCT-based face detection method comprising a. The method of claim 1, In the step (b), the following equation for the NMCT conversion,

(here,

Is

excluding

Neighboring pixels that are in contact with

end

0, if

Is a function representing 2 in the case of 1, otherwise 1

Is the concatenate operator.) NMCT-based face detection method characterized in that the application. The method of claim 1, As a verification process for a face detected simultaneously with the face detection process, (f) calculating the coordinates for the face area of the eyes, nose, and mouth, which are registered as candidates; (g) detecting the calculated face area using a module applied to face detection; (h) comparing and combining the detected eye, nose and mouth facial areas; (i) converting the compared and combined face regions into images of reduced, enlarged and angled; (j) performing verification operation using all modules applied to the face detection, and and performing a cumulative operation of matching and registering candidates after the verification operation. A toy manufactured using the method of claim 1. CPU 10 for controlling data input and output from the components provided on the main board 100, A first storage unit 11 for inputting and outputting data required for various files and operations, A second storage unit 12 for inputting and outputting a system program including an OS for booting the main board 100, and the like; A camera 13 for capturing an external image and converting the image into image data; As a display device for visually representing the image data photographed from the camera is provided a main board 100 including a monitor 14 for displaying an input image and a result image, A controller 70 connected to the main board 100 to control data input / output of components mounted on the FPGA 200; A first memory unit 60 for inputting and outputting filter data and image data stored under the control of the controller 70; A first memory unit controller 61 for controlling input / output data of the first memory unit 60; NMCT-based, characterized in that it comprises an FPGA (200) having a second memory unit 62 for storing intermediate results for face detection, operation and verification by the control of the controller 70 Face detection device. The method of claim 5, An audio input / output unit 18 for converting and outputting an operating state of the main board 100 into an audible signal including voice or converting and storing an externally applied voice signal; A button input unit 19 including a plurality of buttons for turning on / off power applied to the main board 100 or storing and deleting a recognized face or driving the main board 100; Convex lens 25 for guiding the user's face to the center for the camera to shoot, NMCT-based face detection device, characterized in that it comprises a charging adapter jack (23) for supplying power required for driving the main board (100) and the FPGA (200). The method of claim 5, The FPGA board 200 includes a preprocessing module 71 for converting input color image information into a black and white image; A resize module 72 which gradually reduces the image so as to detect faces of various sizes converted into the black and white image; A new MCT module 73 for converting the reduced image into a form for applying a face detection filter; A candidate candidate module 74 for detecting a face by applying a face detection filter to the image converted by the new MCT module; An NMCT-based face detection apparatus comprising a grouping module 75 for outputting a final result by grouping face regions detected by various sizes and positions in the candidate detection module. The method of claim 5, NMCT-based, characterized in that the RS232 interface 16, JTAG interface (16, 66) and expansion connector (17, 67) is provided for the connection of the motherboard 100 and the FPGA (200) Face detection device. A toy manufactured using the apparatus of claim 5.

Images