JP3440641B2 - Operation start position detection method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention recognizes human gestures and hand gestures, interfaces humans with machines based on the recognition results, and detects a start position of motions that can be used for pointing devices and sign language motion recognition. Regarding

[0002]

2. Description of the Related Art In the method of understanding human gestures and hand gestures, when specifying the start position of a motion, the title: "Detection of motion information and application to human interface" (H
C91-33: Yasuhiro Machii et al .: The Institute of Electronics, Information and Communication Engineers, IEICE Technical Report, Human Communication: 1991), a position identification method using coordinates in an arbitrary absolute space has been conventionally used, and as a representative point of the start position. , The detected position of the center of gravity of any object is used as the representative point.

[0003]

When the coordinates of an arbitrary absolute space are used in the prior art, the coordinate value of the specified starting position is significant because the size of the body is different when the subject is different. Moreover, there is a problem that the number of databases will increase in proportion to the number of subjects if there is no database for each subject. Moreover, when only the center of gravity is taken as the representative point of the starting position, even if the subject points at the lips with meaning, if the center of gravity of the palm is taken,
The position of the center of gravity represents the vicinity of the neck, and the position intended by the subject is not the representative point.

Therefore, in the present invention, the space for detecting the start position can be divided by the photographed body characteristics of the subject, and the start motion position can be expressed by the relative position with respect to the subject. Can also specify the starting movement position, and when the subject consciously points to any part of the body, the subject's intention is reflected in the starting movement position by making the tip of the finger the representative point of the palm. The purpose is to be able to.

[0005]

In order to achieve the above object, the present invention provides at least means for photographing the upper half of the body of a subject with a compound-eye camera, means for extracting the body of the subject who has been photographed, and the body of the subject. Along the spatial area dividing means, a means for dividing the area along each part of the face, a means for extracting at least either the left or the right palm, a means for detecting the start timing of the operation, the characteristics of the palm image Means for extracting the palm shape, means for determining whether the palm shape is the designated shape, means for determining whether the palm shape is pointing in the face, and determining whether the representative point of the palm is the center of gravity position or the tip position of the fingertip. It is an operation start position detection method including a means for determining and a means for registering the detected start operation position.

[0006]

In the starting motion position detecting method of the present invention, the database of the same starting motion position can be used even when the subject is changed by dividing the spatial region according to the body feature of the subject. By determining whether the shape of the vertical palm is the designated shape, it is possible to detect the start motion position intended by the subject.

[0007]

Embodiment 1 The operation of one embodiment of the present invention will be described with reference to the flowchart of FIG. As shown in FIG. 2, a person is extracted from the image in which at least the upper half of the person is photographed by subtracting the image of only the background.

(Step 1) The body image extracted in step 1 is binarized by using the discrimination function fg.

[0009]

[Equation 1]

The threshold value is variable because it varies depending on the light source. Next, the contour line of the body is extracted by using an edge extraction filter as shown in (FIG. 3).

A search is performed from the upper part of the image, and the line where the outline of the person and the line parallel to the X-axis are in contact with each other as shown in FIG. Of the lines that are in contact with the HUL and intersect with the outline of the person and the lines parallel to the Y axis, the left side line in the image is the line FRL representing the right side of the face, and the right side line is the line FLL representing the left side of the face. The FRL is extended vertically, and the point intersecting the outline is frlp (Xf, Yf). In addition, a search is performed from the left side of the image, and a point that is parallel to the Y axis and intersects with the outline of the body is found as tempp (Xt, Yt).
The contour line is searched from frlp to tempp, and the point having the maximum curvature is set as the right shoulder point shp. A line passing through shp and parallel to the X-axis is SUL, and a line parallel to the Y-axis is SHRL. The middle line between FRL and FLL is MCL. A line symmetrical to SHRL about the axis of MCL is SHLL, and a line symmetrical to ERL is ELL. The line parallel to the X-axis at the position of 3/4 of SUL and HUL is NEL, and the middle line between SUL and X-axis is BML.

The ERL, SHRL, FRL, MCL,
FLL, SHLL, ELL, HUL, NEL, SUL,
Using BML, the regions of the right camera and left camera images are divided as shown in FIG. Next, as shown in FIG. 5, the intersection point generated by the line segment is obtained. The right and left corresponding points have the same intersection number in the right and left camera images. The coordinate value on the image of intersection number 0 is (Xr0, Yr0), (Xl0, Yl0)
If so, it is substituted into (Equation 3) and the spatial position is calculated.

[0013]

[Equation 2]

Similarly, spatial positions are calculated for all intersections. Based on the result, the spatial domain code (0
~ 24) are defined. The area in front of the person by the distances of MCL and SHRL from the spatial area code (0 to 24) is defined as the spatial area code (25 to 49), and the area further in front is defined as the spatial area code (50 to 74). The space area defined by stores the space coordinate values as a space area code table. This makes it possible to divide the region (face, neck, chest, abdomen, side of face, etc.) along the subject's part as shown in FIG. 6, and associate the spatial region code with the subject's part. Will be shown. This area code can also be created by measuring data from the data glove and specifying from the keyboard.

(Step 2) FIG. 7 shows a template image of an area showing a face. Here, the size of the image is xfac
It is defined as e, yface. Define eye, ear, mouth, cheek, chin, forehead, and nose areas in the template image. The area 11 showing the face of the spatial area code is normalized using the template image. For example, for the mouth area, see FIG.
In the template image of, the start position (xt_m, yt_
m) Let the size be xt_mlen and yt_mlen.

When the size of the space area code 11 indicating the face is the starting point (xf, yf), the height ylen, and the width xlen, normalization is performed. , The start point of the mouth area of the face is (xf_
mouth, yf_mouth), the height is xlen_m
If outh and the width are yflen_mouth, the mouth area of the face is expressed by (Equation 3).

[0017]

[Equation 3]

Similar to the above, the eye, ear, cheek, chin, forehead, and nose areas are calculated, and the face area is divided.

(Step 3) Next, FIG. 8 shows a state in which the skin color information is extracted based on the color information. An object that falls within the threshold value indicating the size of the palm is identified as the palm. The determination of the open palm is expressed by the discriminant function fg (Equation 4).

[0020]

[Equation 4]

Therefore, Lab_1 and Lab_2 shown in FIG.
The palm on the right is determined, and the palm on the right and the palm on the left are specified.

(Step 4) FIG. 9 shows a state in which palms are extracted in time series. At time T (n), the center of gravity of the right camera on the palm of the palm is (Xc1_n_r, Yc1_n_
r), the center of gravity of the left camera is (Xc1_n_l, Yc1
_N_l), the spatial position of the palm is calculated by (Equation 5).

[0023]

[Equation 5]

When the difference is calculated from the position of the center of gravity at time T (n + 1) and the moving distance is d, it is expressed by (Equation 6).

[0025]

[Equation 6]

At least when the moving distance d of the right or left palm exceeds the threshold value, the time T (n + 1) is detected as the start timing.

(Step 5) As shown in (FIG. 10), the characteristics of at least one of the palms extracted in Step 4 are set at the center of gravity (Xc1_r (l), Ycl_r).
(L)), the coordinates of two points indicating the maximum length (Xm12_r
(L), Ym12_r (l)) (Xm11_r (l),
The angle θ between Ym11_r (l)) and the line parallel to the X axis and the line connecting the two points indicating the maximum length is stored in (Table 1).

(Step 6)

[0029]

[Table 1]

Next, as shown in claim 5, the curvature of the contour line of the fresh palm extracted in step 4 is obtained by (Equation 7), and there is a threshold value of curvature. Here, a point exceeding 1 is defined as one point of the polygon. , The palm is approximated to a polygon. Next, as shown in FIG. 11, a convex closed figure of a freshly approximated polygon is created.
The difference of the freshly-approximated polygons is taken from the convex hull figure. At this time, if there is a concave portion, a residual figure remains as shown in FIG. The maximum curvature point of the residual figure is Pc as shown in FIG.
ur. When there is one residual figure, a perpendicular is drawn from the maximum point of curvature to the longest side, and the discriminant function shown in (Equation 8) is used to determine whether or not the shape is a designated shape by using the ratio of the sides.

[0031]

[Equation 7]

[0032]

[Equation 8]

The figure shown in FIG. 12 is determined to be the indicated shape by the discrimination function (Equation 8). (Step 7) When the shape of the fresh palm image extracted in Step 4 is determined to be the designated shape in Step 7, the two points indicating the maximum length are connected among the coordinates of the two points indicating the maximum length obtained in Step 6. A line segment orthogonal to the elliptical line segment is drawn, and one having a smaller length indicating the palm area is detected as the tip position of the pointing shape. In FIG. 10, Xmr1_p and Ymr1_p are detected as the tip positions of the indicated shape.

(Step 8) Coordinates (Xmr1_p, Ymr1) of the tip positions of the palm images of the right camera and the left camera with the spatial position of the tip position of the pointing shape obtained in step 8
_P), (Xml1_p, Yml1_p) to (Equation 9)
The spatial position (Xw_p, Y
w_p, Zw_p) is calculated.

[0035]

[Equation 9]

When the obtained spatial position corresponds to the region (0 to 24) obtained in step 2, it is determined that the body contact position is present. When the spatial position is in the area (25 to 74) obtained in step 2, it is determined that the palm is not in the body contact position.

(Step 9) When it is determined in step 7 that the shape is not the designated shape, or even if the designated shape is not the physical contact position in step 9, the center of gravity position is registered as a representative point of the palm. It

(Step 10) When it is determined in step 9 that the body contact position is present, it is next determined whether or not the tip position is inside the face. Of the spatial areas obtained in step 2, the area in area 11 is determined to be inside the face.

(Step 11) When it is determined in step 11 that the inside of the face is present, the body part is detected as the spatial area 11 of the face obtained in step 2, and the position of the center of gravity of the fresh palm detected in step 9 is the body position. Detect which area it is in.

(Step 12) When it is determined in step 11 that the inside of the face is instructed and contact is made, it is detected which part of the face detected by step 3 the position indicated by the representative point of the palm corresponds to. To do.

(Step 13) When the newly detected palm shape is in the indicated shape and is in contact with the body and is inside the face, there are two kinds of codes, a spatial region code indicating the body part and a code indicating the face part, as the start motion position code. To register.

(Step 14) When the palm shape is other than that in step 14, one type of spatial area code indicating the representative point of the palm is registered as the start operation position code.

(Step 15) By carrying out this embodiment, it is possible to spatially divide the region according to the body characteristics of the subject, and the movement start position can be expressed by the relative position with respect to the subject. When consciously pointing to, the correct starting motion position can be specified by determining the pointing shape at that time.

(Second Embodiment) Instead of step 3 described in the first embodiment, the eye and mouth regions are detected based on the color information of the face region in the extracted body region, and the barycentric position thereof is detected. Ask for. The eye region and the mouth region are determined based on a certain threshold from the position of the center of gravity.

As shown in FIG. 16, which is detected as an example, the center of gravity of the right eye is (xre_cen, yre_ce).
n), the start point, height, and width of the right-eye area are o ″ (xf_reye, yf_reye), xlen, respectively.
If _reye and ylen_reyr, then they are respectively expressed by (Equation 10).

[0046]

[Equation 10]

Similarly, the left eye is also calculated. If the position of the center of gravity of the mouth area is (xm_cen, ym_cen), the start point, height, and width of the mouth area are p ″ (xf
_M, yf_m), m_xlen, and m_ylen are represented by (Equation 11).

[0048]

[Equation 11]

From the above-mentioned positional relationship between the eye area and the mouth area,
Calculate forehead, chin, nose and cheek areas. The ear, forehead, chin, and nose area start positions are q ″ (xf_rear, yf_r
ear), r '' (xf_head, yf_hea
d), t '' (xf_chin, yf_chin),
s ″ (xf_nose, yf_nose) and the widths are xlen_ear, xlen_head, xl, respectively.
en_chin, xlen_nose, and height is yl
en_ear, ylen_head, ylen_chi
If n, ylen_nose, the right ear region is (Equation 1
In 2), the forehead area is (Equation 13), the jaw area is (Equation 14),
The nose area is represented by (Equation 15).

[0050]

[Equation 12]

[0051]

[Equation 13]

[0052]

[Equation 14]

[0053]

[Equation 15]

A region other than the face region is defined as a cheek region. According to this embodiment, at least one face area of the eyes, ears, cheeks, chin, forehead, and nose area is divided. The procedure is the same as that of the first embodiment except step 3 of the first embodiment.

By carrying out the present embodiment, it is possible to extract the eyes and mouth in the face and divide the face into areas according to the subject.

(Embodiment 3) Instead of the start timing of the above step 5 shown in Embodiment 1, the expression expressed by (Equation 16) is used between adjacent frames of the freshly extracted image of the palm, Correlate images. When the correlation value is lower than an arbitrary value, the operation start timing is set in step 5 even if the moving distance of the center of gravity position is within an arbitrary threshold value, and even if the position is the same, the operation is started by the change of the hand shape. You can get the timing. The steps other than step 5 are the same as those in the first embodiment.

[0057]

[Equation 16]

By taking the correlation of the brightness value between the images,
Even when the center of gravity of the palm is not moved, the operation start timing can be set even when the shape changes.

(Embodiment 4) An algorithm described in claim 6 is shown in FIG. 13 instead of step 7 of the embodiment.
The coordinates indicating the maximum length in (Table 1) are connected by a straight line.

(Step a) θ degree affine transformation is performed centering on one of the coordinates showing the maximum length. FIG. 14 shows a diagram after the affine transformation is performed.

(Step b) Y is added to the image after affine transformation.
Draw a line parallel to the axis and measure the length of the part showing the palm of the image. At this time, the number of times the line parallel to the Y axis is inverted is detected.

(Step c) When there is a line whose number of inversions is 3 or more, the above palm shape is determined as a non-designated shape. When the number of inversions is 2, the process proceeds to the module for obtaining the curvature.

(Step d) The curvature of the figure in which the length of the palm of the figure after the affine transformation is measured is calculated, and the point exceeding the threshold value is set as one of the vertices of the polygon, and the polygon is approximated. FIG. 15 shows a polygon-approximated image.

(Step e) Among the vertices of the image shown in FIG. 15, the vertices whose Y coordinates are present within a certain threshold value are grouped in order. In the image of FIG. 15, vertices 1, 2,
3, 4 are in group 1, vertex 5 is in group 2, vertex 6,
7 is registered in group 3, and vertices 8 and 9 are registered in group 4.

(Step f) The lengths of the groups obtained above parallel to the X-axis are xlen1 and xlen, respectively.
2, xlen3, xlen4, and the average of the Y coordinates at that time is high1, high2, high3, high4.
Then, the positional relationship of the groups is stored in (Table 2).

(Step g)

[0067]

[Table 2]

The values in Table 2 are taken out and the discrimination function (Equation 17) is used.
Check to see if it is the indicated shape. The figure after the affine transformation shown in FIG. 15 is determined to be the designated shape.

[0069]

[Equation 17]

(Step h) The same method as in Example 1 is adopted except for Step 7 in Example 1. By using the determination of the pointing shape described above, the correct operation start position can be detected when the body part is specified.

(Embodiment 5) FIG. 1 is a diagram showing an algorithm for determining a designated shape instead of step 7 of Embodiment 1.
Note in 7. FIG. 18 shows a diagram in which the fresh palm image extracted in step 4 has been subjected to the degeneration process any number of times. Next, the spread processing is applied for the number of times of degeneration. FIG. 19 shows a diagram subjected to the spread processing. The difference of the widened image is taken from the original horizontal image. The remaining image is shown in FIG. If the feature of the difference image is taken, the number of labelings is one, and the length is not less than an arbitrary threshold value, the fresh palm image extracted in step 4 shown in the first embodiment is determined as the designated shape.
The same method as in Example 1 is used except for Step 7 in Example 1.

By using the above-described pointing shape determining method, it becomes possible to detect the pointing shape more easily.

[0073]

[Effects of the Invention] Since the spatial region is divided along the body of the subject, it is possible to register at the same start position region even when the subject is different, and the body part of the subject can be registered at the start timing of the manual operation. By detecting the pointing shape, it is possible to specify the starting motion position that the subject himself started.

Prepare one template image of the face area,
It is possible to perform normalization by the actually extracted image, and it is possible to minimize the area of the face part to be registered.

With the color information of the extracted face area of the subject himself / herself, the face part can be specified by calculation, it is not necessary to register the template image, and the face area can be divided according to the subject.

By correlating the images between the frames of the palm area, the position where the palm shape changes even when the palm is not moving can be used as the operation start timing.

By determining the shape in which the representative point of the palm is pointing to a part of the body, it is possible to accurately specify the starting operation position.

[Brief description of drawings]

FIG. 1 is a flowchart showing an algorithm of a first embodiment.

FIG. 2 is a diagram illustrating a background difference image in which a person is extracted according to the first embodiment.

FIG. 3 is a diagram showing body contour lines shown in the first embodiment.

FIG. 4 is a diagram showing a body detection line shown in the first embodiment.

FIG. 5 is a diagram showing a region division diagram and intersections according to the first embodiment.

FIG. 6 is a diagram showing that the spatial region division according to the first embodiment is performed.

FIG. 7 is a diagram showing a template image of each face part according to the first embodiment.

FIG. 8 is a diagram showing a state in which skin color information of the first embodiment is extracted.

FIG. 9 is a diagram showing a state in which a palm is extracted in a time series of the first embodiment.

FIG. 10 is a diagram showing that the features of the palm of the first embodiment have been detected.

FIG. 11 is a polygon-approximated palm of the first embodiment +
Figure showing the palm convex shape

FIG. 12 is a diagram showing a residual graphic according to the first embodiment.

FIG. 13 is a flowchart of the pointing shape determination algorithm of the fourth embodiment.

FIG. 14 is a diagram showing a palm figure after affine transformation according to a fourth embodiment.

FIG. 15 is a diagram showing a polygon-approximated figure after the affine transformation of the fourth embodiment.

FIG. 16 is a diagram showing a face part extraction image of the second embodiment.

FIG. 17 is a flowchart of the algorithm of the fifth embodiment.

FIG. 18 is a diagram showing a retracted palm of the fifth embodiment.

FIG. 19 is a diagram showing that the degenerated image of the fifth embodiment has been subjected to spread processing.

FIG. 20 is a diagram showing an image obtained by subtracting the spread image from the palm image of the fifth embodiment.

[Explanation of symbols]

Xw, Yw, Zw Spatial position coordinates (Xr0, Yr0) of the intersection point of the right camera (Xl0, Yl0) Coordinates of the intersection point number 0 of the left camera ERL (ElbowRightLine) Line showing the right elbow SHRL (ShoulderRightLine) Line showing right shoulder FRL (FaceRightLine) line showing right side of face MCL (MiddleCenterLine) line showing center of body FLL (FaceLeftLine) line showing left side of face SHLL (ShoulderLeftLine) line showing left side of shoulder ELL left (ELL) Line indicating the elbow SUL (ShoulderUpperLine) Line indicating the upper part of the shoulder shp (ShoulderPoint) Point indicating the position of the shoulder frlp (FaceRightLineP) int)
A line that intersects the contour line that extends FRL vertically. Tempp (TemporaryPoint) The line that the body first touches the line parallel to the Y axis. Xface The length of the face area in the template image in the X direction. The face area in the yface template image. In the Y direction (xt_m, yt_m) Start position of the mouth area in the template image xt_mlen Length of the mouth area in the X direction in the template image yt_mlen Length of the mouth area in the Y direction in the template image (xf, yf ) Start point xlen of the face area in the extracted image Length of the face area in the xlen direction in the extracted image ylen Length of the face area in the ylen direction in the extracted image (xf_mouth, yf_mouth) Coordinates indicating the start point of the mouth area of the extracted image xlen_mout Extraction of the length of the mouth area of the extracted image in the X direction ylen_mouth Length of the mouth area of the image in the Y direction o'o '' right eye area start address p, p ', p''mouth area start address q', q '' right ear area start address r, r ', r '' forehead area start address s ', s''nose area start address t', t '' jaw area start address Lab_1, Lab_2 right, left palm area Lab_3 face area (Xc1_n_r, Yc1_n_r) right camera Center position (Xc1_n_l, Yc1_n_l) indicating the right palm of the nth frame of the left camera Center of gravity position (Xc1_n + 1_r, Yc1_n + 1_r) indicating the right palm of the nth frame of the left camera (Xc1_n + 1_l_l_l, left) Center of gravity position (Xm11_r (l), Ym11_r (l)) indicating the left palm of the camera n + 1 frame Right (left Maximum point indicating the length coordinates of the palm of the camera (Xm12_r (l), Ym12_r (l)) Right (Left) camera maximum point indicating the length coordinates of the palm (Xw_p, Yw_p, Zw_p) palm tip spatial coordinates of

Continuation of the front page (56) References JP-A-2-144675 (JP, A) JP-A-7-282235 (JP, A) JP-A-8-320920 (JP, A) Matsuo Hideaki Outside 3 people, non-contact Sign Language Motion Recognition Algorithm, Human Interface News and Report, The Society of Instrument and Control Engineers, January 23, 1995, Vol. 10, No. 1, pp. 41-46 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06T 1/00 G09B 21/00 JST Plus file (JOIS)

Claims (7)

(57) [Claims] 1. In the motion start position detection of a sign language motion, by subtracting at least the image of a subject whose upper body is imaged by a compound-eye camera and an image of a background showing the same place where the subject is not photographed, A step of extracting the body of the subject, a step of dividing the spatial region along the extracted body part of the subject, a step of dividing the region showing the face of the spatial region into a specific region, at least the right, A step of extracting an image of one of the left palms, a center of gravity of the image of the palm of the left is detected, and a step of detecting as a motion start timing when the barycentric position is displaced between frames that are temporally behind, and Storing the barycentric position of the palm image and the coordinates of two points indicating the maximum length as the tip candidate position; A step of determining whether the shape is a pointing shape that specifies a part of the body, a step of setting the center of gravity as a representative point of the palm if not the pointing shape, and when the shape of the palm image is the pointing shape, the The step of setting the point having a small inner hand area of two tip position candidates as the tip position of the palm, and setting the representative point of the palm as a representative point of the palm is an area showing the face of the subject, When the face area is not shown, a step of specifying where the tip position is in the divided space area and registering one kind of motion start position code; and a representative point of the tip position is the subject's face. When indicating a region indicating a region, the tip position is the eyes, ears, lips of the face region,
A method for detecting a motion start position of a sign language motion, comprising the steps of specifying where on the cheek, forehead, chin, and nose, specifying the body part in the divided space as a face, and registering two types of motion start position codes. 2. A step of dividing a region showing a face in a spatial region into specific regions includes preparing a space of a template face region modeling at least one region of eyes, ears, lips, cheeks, foreheads, chins and noses. The operation start position detecting method according to claim 1, wherein the space area of the face is divided by enlarging and reducing the template image according to the size of the space area of the face. 3. The step of dividing the area showing the face of the spatial area into specific areas extracts the eyes and lips based on the color information of the area showing the face of the spatial area, and based on the positional relationship between the eyes and the lips, Ears, chin, cheeks,
The operation start position detecting method according to claim 1, wherein at least one area of a nose and a forehead is shown, and the space area of the face is divided. 4. The operation start position according to claim 1, wherein the step of detecting the operation start timing takes an autocorrelation between consecutive frames of all the palm image extraction areas, and sets the operation start timing when an arbitrary threshold value is exceeded. Detection method. 5. The curvature of the outer contour line of the palm image obtained by the step of determining whether the shape is a pointing shape that specifies a part of the body,
The palm is approximated to a polygon with a point exceeding a certain threshold as one point of the polygon, and the corners of the polygon are classified into a convex portion and a concave portion, and when the concave portion has only one or two corners, the instruction The operation start position detecting method according to claim 1, wherein the method is determined as a shape. 6. The step of determining whether it is a pointing shape that specifies a part of the body is a straight line passing through the coordinates of two points indicating the maximum length of the palm image and X.
The angle with respect to the axis is calculated, and the palm image is affine-transformed by the angle around one of the two points indicating the maximum length, and Y of the palm image after the affine transformation is calculated.
The projection of the axis is performed, the curvature of the projected image is calculated for the Y axis, the points exceeding a certain threshold are detected as the vertices of the polygon, the polygon approximation is performed, and the value of the Y axis is a group within an arbitrary threshold. 2. The operation start detection method according to claim 1, wherein the shape of the palm image is determined as the designated shape based on the relationship between the groups. 7. The step of determining whether the shape is a pointing shape that specifies a part of the body is subjected to degeneration processing of the palm image for any number of times to create a degenerate image of the palm image, and the degenerated image of the palm image is the arbitrary image. The restored image that has been subjected to the diffusion process for the number of times is subtracted from the palm image to create a palm difference image, and the palm difference image is labeled. The number of labels is one and the length is equal to or greater than an arbitrary threshold value. The operation start detecting method according to claim 1, wherein at the time of, the shape of the palm image is determined as the designated shape.