JPH05215531A - Three-dimensional-body identifying and processing method - Google Patents

Abstract

PURPOSE:To extract more stabler characteristics by identifying a body by using the three- dimensional information such as the normal lines of the surface of a face as the characteristics, which are not subjected to the change in observing point. CONSTITUTION:An input means 1 performs the adeguate conversion of the three-dimensional shape data of a person's face by the three-dimensional measurement and sends the result into a reference-point extracting means 2. The means 2 investigates the curvature of the surface of the face and extracts the nose-top point, the ear-hole points and the like, which indicate the maximum values, as the reference points. A reference-attitude computing means 3 computes the middle point between the right and left ear-hole points as the reference position and the vector connecting the reference position and the nose-top point as the reference direction. The values are converted into the X-Y-Z coordinate systems with a reference-attitude converting means 4. Then, the position of a collating region is computed with a collating- region-position computing means 5. A required region is cut out with a region cutting-out means 6. Characteristic extraction processing is performed for the data with a characteristic extracting means 7, and a collating pattern is formed. In a collating means 8, the pattern is collated with a standard pattern in an identifying dictionary file 9, and the similarity scale between both patterns is made to be the numeric value. The value is judged with a judging means 10, and the result is outputted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a three-dimensional object identification process for identifying a class to which a plurality of identification target categories are defined by different shapes by using the surface shape information. It is about the method.

In the following, a person's face will be mainly taken up as an example of an object to be identified having the above-mentioned properties, and an application to personal identification based on a face shape will be described as an example. It goes without saying that this method is widely applicable to identification of various three-dimensional objects.

[0003]

2. Description of the Related Art Conventionally, in identifying a person's face, information obtained from a two-dimensional grayscale image of the face has been used as a feature. First, there is a method that features a grayscale image itself. For example, "Kosugi Shin," Facial Image Identification and Feature Extraction Using Neural Networks ", Jitsugaku Giho, Vol.91, No.56,
pp9-16 ". In this, the face is identified by using a normalized two-dimensional grayscale image of the face area.
There is also a method that features a two-dimensional position / shape such as a face feature obtained from a two-dimensional grayscale image. For example, "Eiichi Hagiwara, Isao Masuda," Personal ID based on facial images mainly for pattern matching ", IEICE Technical Report, Vol.88, No.112, pp.
53-60 ". In this, the face is identified based on the feature length of the face obtained from the two-dimensional grayscale image.

However, these methods are easily affected by the direction and inclination of the face, illumination, etc., and it is not possible to obtain a constant two-dimensional grayscale image. Furthermore, information such as unevenness and depth of the face cannot be used sufficiently, which is insufficient for expressing a subtle shape of the face. As a result, there is a problem that identification cannot be performed well.

On the other hand, as a feature, there is a method of using a distance image obtained from the three-dimensional shape information of the face. For example, "Nobuhiko Masui, Shigeru Akamatsu, Yasuhito Suenaga," Fundamental Study of Face Image Recognition by 3D Measurement ", TV Gakugiho, Vol.14, No.36,
pp7-12 ”. This is an approach that focuses more directly on the distribution of the three-dimensional shape of the face surface, and identifies the face using the distance image in the front face image, which is the feature that represents the overall three-dimensional shape of the face. There is.
Therefore, in this method, facial features such as unevenness and depth of the face can be well expressed without being affected by illumination.

[0006]

However, even this method is affected by the variation of the input condition such as the orientation / tilt of the face, that is, the posture of the human head, and therefore, the stability of the feature identifying ability is still unsatisfactory. It is enough. As a result, there is a problem that identification cannot be performed well.

According to the present invention, for example, in a method of focusing on a distance image of a face, which is a feature representing a global three-dimensional shape of a face, a method of determining a reference direction of a face for creating a distance image, that is, a normal viewpoint is used. The purpose is to solve the problem of how to accurately perform the conversion.

[0008]

In order to solve the above-mentioned problems, as a feature, three-dimensional information of a face that is invariant to changes in input conditions is used. That is,
As a feature that is not affected by changes in the viewpoint, three-dimensional information called the normal line of the face surface is used for identification.

[0009]

According to the above-mentioned means, when performing personal identification using the human face shape, the three-dimensional shape is not used as the characteristic for identification, instead of using the conventionally used two-dimensional grayscale image or range image. It becomes possible to extract more stable features by using the three-dimensional information called the normal line obtained from the above, and moreover, it is possible to express a subtle three-dimensional shape of the face, and it is possible to perform stable individual identification by the face shape. .. Moreover, since the features are converted from the three-dimensional shape data to the normal line data, the amount of information can be compressed.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be specifically described below with reference to the drawings. In all the drawings for explaining the embodiments, parts having the same function are designated by the same reference numerals, and repeated description thereof will be omitted.

FIG. 1 is a block diagram showing a schematic configuration of an embodiment of a three-dimensional object identifying method of the present invention, and FIG.
It is a block diagram which shows the structure of the functional system of the feature extraction means of FIG.

In FIG. 1, 1 is input means, 2 is reference point extraction means, 3 is reference posture calculation means, 4 is reference posture conversion means, 5 is collation area position calculation means, 6 is collation area cutout means, and 7 is Feature extraction means, 8 is collation processing means, 9 is an identification dictionary file, 10 is determination processing means, and 11 is control means for managing the progress of the entire processing.

The reference point extraction means 2, the reference attitude calculation means 3, ..., And the control means 11 are constituent elements that can be built in the same computer. Feature extraction means 7
As shown in FIG. 2, is composed of a feature amount conversion unit 71 and a normal distribution extraction unit 72.

In the input means 1, the three-dimensional measuring device is used to input the three-dimensional shape data of the person's face, the data is converted into a format suitable for the subsequent processing, and the reference point extracting means 2 is used. Is sent to the reference attitude conversion means 4.

A method for measuring the surface shape of a three-dimensional object is described in, for example, "Seiji Iguchi," Current State of Three-dimensional Sensing Technology ", Image Lab, Vol.1, No.4, pp44-47". Any method can be used.

The reference point extracting means 2 extracts a reference point for calculating a reference posture and a reference point for calculating a matching area based on the three-dimensional shape data. The face reference points are extracted by, for example, examining changes in the curvature of the face surface. Specifically, near the center of the front of the face, the point where the absolute value of curvature is maximum is the apex of the nose, and near the center of the side of the face, the point where the absolute value of curvature is maximum is the ear hole point. , Is extracted as.

The reference attitude calculating means 3 calculates the reference attitude using the reference points obtained by the reference point extracting means 2. The reference posture is calculated, for example, as the reference position and the reference direction based on the reference points (the apex of the nose and the left and right ear hole points) as shown in FIG. Specifically, the midpoint C of the left and right ear hole points L and R is used as a reference position, and the vector CN connecting the mid point of the left and right ear hole points and the nose vertex N is used as the reference direction.

The reference attitude conversion means 4 creates reference attitude data for the data input by the input means 1 using the reference attitude calculated by the reference attitude calculation means 3. The conversion to the reference orientation is performed by, for example, moving the reference position C so that it coincides with the origin of the XYZ coordinate system, and then changing the reference direction vector CN.
Rotates so as to coincide with the Z axis positive direction. This is the reference attitude data.

The matching area position calculating means 5 uses the reference points obtained by the reference point extracting means 2 for the reference attitude data created by the reference attitude converting means 4 and uses the matching areas required for matching. The position is calculated.

The position of the matching area is calculated, for example, based on the reference points (nasal vertex and left and right ear hole points) as shown in FIG. Specifically, the procedure is as follows. (1) Obtain the average M of the distances between the apex of the nose and the left and right ear hole points. (2) An area in which the distance from the apex of the nose is a constant multiple of M (αM) is set as the matching area.

In the collation area cut-out means 6, an area required for collation with respect to the reference orientation data created by the reference orientation conversion means 4 according to the position information of the collation area calculated by the collation area position calculation means 5. Is cut out.

In the feature extracting means 7, for the three-dimensional shape data for collation cut out by the collation area cutting means 6,
The extraction process of the features to be compared at the time of matching is performed, and the matching pattern is generated.

The processing here will be described with reference to FIG. The three-dimensional shape data sent to the feature extraction means 7 approximates a curved surface to a plane with an arbitrary size (of course, a local curved surface is possible) in the feature amount conversion unit 71, and uses it as a unit normal vector of the plane. The area is required. To approximate a curved surface to a plane,
It is obtained by using at least three points on the curved surface to find a plane passing through the three points, and approximating the curved surface to the plane. The approximated “unit normal vector of plane” is given in consideration of the normal line in the approximated plane.
From the result, the normal distribution extraction unit 72 obtains a normal distribution in which the size of each unit normal vector is represented by the area of the plane, that is, a set of vectors having a size proportional to the area. The three-dimensional shape data is feature-converted into a normal distribution and sent to the matching processing means 8.

Next, in the collation processing means 8, the feature pattern consisting of the features extracted by the feature extracting means 7 is subjected to the processing up to the feature extracting means 7 and registered in advance so that the prepared identification dictionary is prepared. By collating with the standard pattern in the file 9, the similarity measure between the two is digitized.

In the judgment processing means 10, a threshold value having an optimum value for the form in which the data group of the similarity measure between the input pattern calculated by the matching processing means 8 and the standard pattern of each category is to be used It is determined by processing and the result is output.

As can be seen from the above description, according to the present invention, when performing personal identification using a human face shape, three-dimensional shape information, which is a normal to the face surface, is used instead of using a two-dimensional grayscale image. Since it is used, it is not affected by the orientation / tilt of the face, lighting, etc., and information such as the unevenness and depth of the face can be used sufficiently, and a delicate shape of the face can be expressed, so a stable feature Can be extracted, and individual identification by face shape can be performed stably. Further, since the features are converted into normal line data instead of using the three-dimensional shape data as they are, the amount of information can be compressed.

The present invention has been specifically described above based on the embodiment of the identification system mainly for identifying the face of a person, but the present invention is not limited to the above-mentioned embodiment. Needless to say, various changes can be made without departing from the gist of the invention.

[0028]

As described above, according to the present invention, since the face data is characterized by the three-dimensional shape information, stable feature extraction can be performed and the three-dimensional shape of the face can be expressed. , It is possible to stably perform personal identification by face shape. Further, since the features are converted into normal line data instead of using the three-dimensional shape data as they are, the data amount can be compressed.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of an embodiment of a three-dimensional object identification method of the present invention.

FIG. 2 is a block diagram showing the configuration of a functional system of the feature extraction means of FIG.

FIG. 3 is a diagram showing an example of a method for obtaining a reference posture of a face.

FIG. 4 is a diagram showing an example of a method of calculating a face matching area.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 input means 2 reference point extraction means 3 reference attitude calculation means 4 reference attitude conversion means 5 collation area position calculation means 6 collation area cutout means 7 feature extraction means 8 collation processing means 9 identification dictionary file 10 determination processing means 11 control means 71 features Quantity conversion unit 72 Normal distribution extraction unit

Claims (2)

[Claims] 1. Input means for inputting the surface shape of a target three-dimensional object by a three-dimensional measuring device, and extraction of a reference point for obtaining a reference attitude and a matching area of the input three-dimensional object. A reference point extracting means for performing the reference attitude, a reference attitude calculating means for obtaining the reference attitude of the three-dimensional object based on the extracted reference points, a reference attitude converting means for moving the input three-dimensional object to the reference attitude, and the extracted Collation area position calculation means for calculating the position of the collation area based on the reference point, and collation area cutout means for performing the clipping operation on the reference posture three-dimensional object based on the calculation result to cut out the area required for collation. , A feature extraction means for performing a process of extracting a feature pattern required for a match from the match region, and a match processing means for performing a match process of the extracted feature pattern and a standard pattern prepared in advance, A determination process means for matching result to determine whether it is appropriate, and control means for controlling contact the respective processing unit, identification processing method of a three-dimensional object which is characterized in that the identification of the three-dimensional object. 2. The feature extracting means approximates a curved surface to a plane with an arbitrary size and obtains a unit normal vector and its area of the plane, and a feature quantity conversion unit, and based on the result, a normal distribution. The three-dimensional object identification processing method according to claim 1, wherein the three-dimensional object identification processing unit is configured with a normal distribution extracting unit to be obtained.