Face detection apparatus, face detection method, and program

A face detection apparatus detecting a face from an image which is captured by an imaging unit and includes the face, the apparatus comprising:

• at least one non-transitory computer readable medium operable to store program code;
• at least one processor operable to read said program code and operate as instructed by the program code, the program code comprising:
• a position detection unit that detects a position of a face part of the face from the image;
• an initial state determination unit that determines a model initial state of a face model on the basis of the position of the face part;
• and
• a model fitting unit that generates a model of the face on the basis of the image by using the model initial state, the model being generated by fitting the face model to the image captured by the imaging unit,
• wherein the face part includes left and right eyes and a nose,
• wherein the initial state determination unit calculates the angle of the face including a pitch angle, a yaw angle, and a roll angle from eye positions of the left and right eyes and a nose position, and determines face positions on the basis of the eye positions and the nose position,
• wherein the face positions include a face position in a plane direction and a face position in a depth direction in an image,
• wherein the initial state determination unit determines the face position in the plane direction on the basis of the nose position and the face position in the depth direction on the basis of a distance between the left eye and the right eye, and determines the model initial state on the basis of the angle of the face and the face positions, and
• wherein the initial state determination unit determines a position of the face in a depth direction by calculating a distance between the left and right eyes from the detected eye position and calculating a ratio between the distance between the left and right eyes and an average distance between the eyes in a standard face which is statistically acquired in advance, and determines the model initial state by enlarging or reducing the model using the position of the face in the depth direction,
• the initial state determination unit acquires a temporary model and rotates the temporary model by using the face angles and moves the temporary model in the plane direction of the image by using the face position in the plane direction and enlarges or reduces the temporary model by using the face position in the depth direction and determines the face model by modifying the temporary model,
• the initial state determination unit calculates the pitch angle based on the following Equation (1):
• θ
• ⁢
• ⁢
• x
• =
• arccos
• ⁡
• (
• dh
• dh
• ⁢
• ⁢
• 0
• ×
• (
• dw
• ⁢
• /
• ⁢
• dw
• ⁢
• ⁢
• 0
• )
• )
• =
• arccos
• ⁡
• (
• R
• ⁢
• ⁢
• 0
• ×
• dh
• dw
• )
• (
• 1
• )
• wherein a distance between the left and right eyes is set to dw, a distance between the nose position and a straight line connecting the left and right eyes is set to dh, a dw and a dh of a standard face statistically acquired in advance is set to dw0 and dh0, respectively, and the pitch angle is expressed as θx;
• the initial state determination unit calculates the yaw angle based on the following Equation (2):
• θ
• ⁢
• ⁢
• y
• =
• arcsin
• ⁡
• (
• dw
• ⁢
• ⁢
• 1
• -
• dw
• ⁢
• ⁢
• 2
• dw
• ⁢
• ⁢
• 1
• +
• dw
• ⁢
• ⁢
• 2
• )
• (
• 2
• )
• wherein a perpendicular line is drawn from the nose position to the straight line connecting the left and right eyes, a distance from an intersection between the straight line and the perpendicular line to the left eye position is set to dw1, a distance to the right eye position therefrom is set to dw2, and the yaw angle is expressed as θy;
• and
• the initial state determination unit calculates the roll angle based on the following Equation (3):
• θ
• ⁢
• ⁢
• z
• =
• arctan
• ⁡
• (
• dy
• dx
• )
• (
• 3
• )
• wherein a value obtained by subtracting an x coordinate of the left eye position from an x coordinate of the right eye position is set to dx, a value obtained from subtracting a y coordinate of the left eye position from a y coordinate of the right eye position is set to dy, and the roll angle is expressed as θz.