CN105260706A - Head gesture detection method based on image comparison and heading gesture system - Google Patents

Abstract

The present invention relates to a head attitude detection method based on image comparison and heading attitude system. The present invention records and establishes a large number of face attitude data lists including face images and attitude information corresponding to the face images through the heading attitude system. When it is necessary to estimate the pose information of the face image, use the face recognition technology to compare the face image to be tested with each face image in the face pose data list one by one, and find the most similar face image to the face image to be tested. For the face image, the pose information of the closest face image is the pose information of the face image to be tested. The present invention establishes a face pose data list, reduces data errors during face image recognition and detection; solves the problem of correspondence between face images and pose information, improves detection accuracy and real-time performance; and can be applied to automatic identity recognition , video surveillance systems and other fields.

Description

A head attitude detection method based on image comparison and heading attitude system

technical field

The invention relates to a head attitude detection method based on image comparison and heading attitude system, which belongs to the intersection field of computer vision and human-computer interaction.

Background technique

In the traditional field of computer vision, such as face detection, face recognition, expression recognition, etc., methods based on frontal face detection and recognition have developed rapidly and have high recognition and detection rates. However, due to the diversity of detected target poses, the performance effect often drops sharply or even fails. Therefore, as a key technology of face recognition, face pose estimation has received great attention and rapid development. At present, the existing face pose detection methods can be roughly divided into two categories: the first is a learning method based on 2D face appearance, which tries to establish a mapping relationship between 2D face appearance features and 3D face pose , and then based on the learning method of 2D face appearance to realize the detection and estimation of pose; the second type is to extract 3D features, or use multiple images from different perspectives to reconstruct the 3D model of the face in 3D space to realize pose detection detection. However, due to the inaccuracy of the pose information of the samples used for comparison and the lack of sample numbers, great errors will be caused to the final detection results.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention provides a head attitude detection method based on image comparison and heading attitude system;

The head posture detection method of the present invention is applied to static images and real-time video sequences, and is suitable for automatic identification, video surveillance systems, computer games, virtual reality and driver fatigue detection systems, etc.

Terminology Explanation

1. Yaw axis, Pitch axis, Roll axis: If there is a person standing at (0, 0, 0) point, facing the positive direction of X axis, head

The top-up direction is the positive direction of the Y-axis, and the right-hand direction is the positive direction of the Z-axis. Then the calculation method of the rotation angle and direction is as follows:

Yaw rotates around the Y axis. The person standing at (0, 0, 0) is the XOZ plane. A positive angle is used as a parameter to turn left, and a negative angle is used as a parameter to turn right.

Pitch is a rotation around the X axis. The person standing at (0,0,0) is the XOY plane. If the parameter is positive angle, it will fall to the right, and if the parameter is negative angle, it will fall to the left.

Roll rotates around the Z axis. The person standing at (0, 0, 0) is on the YOZ plane. If a positive angle is used as a parameter, it will fall backwards, and if a negative angle is used as a parameter, it will fall forward. Taking the human face as an example, pitch is to touch the left and right shoulders with the head; roll is to touch the collarbone with the head and raise the head;

2. Heading and attitude system (heading and attitude system): It is a flight instrument that measures, displays and provides aircraft heading angle and attitude angle signals. This system mainly consists of a full-attitude gyroscope, a magnetic induction sensor or astronomical compass, and a full-attitude indicator. It can indicate to the flight crew the heading angle required for navigation and the roll angle and pitch angle required for driving, and can also provide a unified heading angle and Attitude angle signal.

3. Histogram equalization: refers to the method of adjusting the contrast by using the image histogram in the field of image processing.

Technical scheme of the present invention is:

A head attitude detection method based on image comparison and heading attitude system, the specific steps include:

A. Through the new architecture, establish a face pose database of M measured objects

(1) Establish a face pose database for the measured object m, m=0;

(2) Take the head center of the measured object m as the coordinate origin, and establish a rotation plane with [-π, π] of the Roll axis, Yaw axis, and Pitch axis respectively;

(3) Fix the heading and attitude system on the front of the head of the measured object m, and detect the face attitude in real time; ensure the real-time and accuracy of the attitude information.

(4) Capture the front face image of the measured object m through the heading and attitude system, and record the attitude information A ₀ [a ₀ , b ₀ , c ₀ ] of the front face image of the measured object m, where a ₀ refers to the The angle between the front face of the measured object m and the Roll axis, b ₀ refers to the angle between the front face of the measured object m and the Yaw axis, c ₀ refers to the angle between the front face of the measured object m and the Pitch axis angle; the value ranges of a ₀ , b ₀ , and c ₀ are all (0-0.5)°;

(5) With a rotation range of 1°-2°, within the allowable rotation range of the head of the measured object m, the head of the measured object m rotates on the Roll axis, Yaw axis, and Pitch axis, and a large number of people are photographed in real time. face image, and record the pose information A _i [a _i , b _i , c _i ] corresponding to the face image in real time, and use m as the keyword to enter the face pose data list; a _i refers to the measured position under the current rotation angle The angle between the face of the object m and the Roll axis, b _i refers to the angle between the face of the measured object m and the Yaw axis at the current rotation angle, and c _i refers to the angle between the face of the measured object m and the Yaw axis at the current rotation angle The included angle of the Pitch axis;

(6) When m<M, replace the face, add 1 to m, repeat steps (2)-(5); when m=M, enter step B;

B. Perform face recognition on the face image to be tested, find out the closest face image in the face pose data list, and obtain the pose information of the face image to be tested

a. Preprocessing the human face image to be tested, the preprocessing includes successively reducing the size, histogram equalization, and skin color edge extraction; when the human face image to be tested is a static image, enter step b; when the human face to be tested When the image is a video sequence, enter step c;

After preprocessing, the influence of illumination and identity information on the detection results is reduced.

b. Through face recognition technology, compare the face image to be tested with each face image in the face pose data list one by one, find out the face image closest to the face image to be tested, the closest The pose information of the face image is the pose information of the face image to be tested;

c. Execute step b for each frame of the human face image to be tested in the video sequence to obtain the pose information of each frame of the human face image to be tested, and calculate the average pose information, that is, the pose information of the video sequence.

Preferably according to the invention, a ₀ =0°, b ₀ =0°, c ₀ =0°.

Preferably according to the present invention, the face recognition technology is template matching or eigenface.

The beneficial effects of the present invention are:

1. The present invention solves the problem of face image pose information estimation in static images, video sequences and real-time images;

2. The present invention solves the problem of correspondence between face images and posture information, and improves detection accuracy;

3. The present invention establishes a face pose data list, which reduces data errors during face image recognition and detection;

4. The face posture data list information of the present invention is open source, and data can be added at any time according to needs, and can be applied to fields such as automatic identification and video surveillance systems.

detailed description

The present invention is further limited below in conjunction with embodiment, but is not limited thereto.

Example 1

A head attitude detection method based on image comparison and heading attitude system, the specific steps include:

A. Through the new architecture, establish a face pose database of M measured objects

(1) Establish a face pose database for the measured object m, m=0;

(2) Take the head center of the measured object m as the coordinate origin, and establish a rotation plane with [-π, π] of the Roll axis, Yaw axis, and Pitch axis respectively;

(3) Fix the heading and attitude system on the front of the head of the measured object m, and detect the face attitude in real time; ensure the real-time and accuracy of the attitude information.

(4) Capture the front face image of the measured object m through the heading and attitude system, and record the attitude information A ₀ [a ₀ , b ₀ , c ₀ ] of the front face image of the measured object m, where a ₀ refers to the The angle between the front face of the measured object m and the Roll axis, b ₀ refers to the angle between the front face of the measured object m and the Yaw axis, c ₀ refers to the angle between the front face of the measured object m and the Pitch axis angle; the values of a ₀ , b ₀ , and c ₀ are all 0°;

(5) With 1° as the rotation range, within the allowable rotation range of the head of the measured object m, the head of the measured object m is rotated on the Roll axis, the Yaw axis, and the Pitch axis, and a large number of face images are captured in real time. And record the posture information A _i [a _i , _{bi , c i ]} corresponding to the face image in real time, with m as the keyword, enter the face posture data list; a _i refers to the measured object m under the current rotation angle The angle between the face and the Roll axis, b _i refers to the angle between the face of the measured object m and the Yaw axis at the current rotation angle, and c _i refers to the angle between the face of the measured object m and the Pitch axis at the current rotation angle Angle;

(7) When m<M, replace the face, add 1 to m, repeat steps (2)-(5); when m=M, enter step B;

B. Perform face recognition on the face image to be tested, find out the closest face image in the face pose data list, and obtain the pose information of the face image to be tested

a. Preprocessing the human face image to be tested, the preprocessing includes successively reducing the size, histogram equalization, and skin color edge extraction; when the human face image to be tested is a static image, enter step b; when the human face to be tested When the image is a video sequence, enter step c;

After preprocessing, the influence of illumination and identity information on the detection results is reduced.

b. Through face recognition technology, compare the face image to be tested with each face image in the face pose data list one by one, find out the face image closest to the face image to be tested, the closest The pose information of the face image is the pose information of the face image to be tested;

c. Execute step b for each frame of the human face image to be tested in the video sequence to obtain the pose information of each frame of the human face image to be tested, and calculate the average pose information, that is, the pose information of the video sequence.

The face recognition technology is template matching.

Example 2

According to the head attitude detection method based on image comparison and heading attitude system described in Embodiment 1, the difference is that in the step (5), the rotation range is 2°, and the head of the measured object m is Within the allowable rotation range, the head of the measured object m rotates on the Roll axis, Yaw axis, and Pitch axis, and a large number of face images are captured in real time, and the attitude information A _i [a _i ,b corresponding to the face images is recorded in real time _i , _ci ], with m as the key word, entered into the face pose data list. The face recognition technology is eigenface.

Example 3

According to a head attitude detection method based on image comparison and heading attitude system described in Embodiment 1, the difference is that in the step (5), the rotation range is 1.5°, and the head of the measured object m Within the allowable rotation range, the head of the measured object m rotates on the Roll axis, Yaw axis, and Pitch axis, and a large number of face images are captured in real time, and the attitude information A _i [a _i ,b corresponding to the face images is recorded in real time _i , _ci ], with m as the key word, entered into the face pose data list. The face recognition technology is eigenface.

Claims (3)

1. based on image ratio to the head pose detection method with heading and attitude system, it is characterized in that, concrete steps comprise:

A, by new architecture, set up the human face posture database of M measurand

(1) human face posture database is set up to measurand m, m=0;

(2) with the number of people center of measurand m for true origin, respectively with Roll axle, Yaw axle, Pitch axle [-π, π] set up Plane of rotation;

(3) heading and attitude system is fixed on the number of people front of measurand m, human face posture is detected in real time;

(4) by the front face image of heading and attitude system shooting measurand m, the attitude information A of record measurand m front face image ₀[a ₀, b ₀, c ₀], wherein, a ₀refer to the front face of measurand m and the angle of Roll axle, b ₀refer to the front face of measurand m and the angle of Yaw axle, c ₀refer to the front face of measurand m and the angle of Pitch axle; a ₀, b ₀, c ₀span be (0-0.5) °;

(5) with 1 °-2 ° for rotational steps, in the scope that measurand m head can allow rotation, the head of measurand m rotates at Roll axle, Yaw axle, Pitch axle, a large amount of facial image of captured in real-time, and the attitude information A that real time record facial image is corresponding _i[a _i, b _i, c _i], take m as keyword, in the data list of typing human face posture; a _irefer to the face of measurand m and the angle of Roll axle under present rotation angel degree, b _irefer to the face of measurand m and the angle of Yaw axle under present rotation angel degree, c _irefer to the face of measurand m and the angle of Pitch axle under present rotation angel degree;

(6) as m < M, change face, m adds 1, repeats step (2)-(5); As m=M, enter step B;

B, recognition of face is carried out to facial image to be measured, find out facial image the most close in human face posture data list, draw the attitude information of facial image to be measured

A, carry out pre-service to facial image to be measured, described pre-service comprises carries out that size reduces, histogram equalization, colour of skin edge extracting successively; When facial image to be measured is still image, enter step b; When facial image to be measured is video sequence, enter step c;

B, by face recognition technology, each facial image in facial image to be measured and human face posture data list to be compared one by one, find out the facial image the most close with facial image to be measured, the attitude information of this most close facial image and the attitude information of facial image to be measured;

C, facial image to be measured for each frame of video sequence is performed step b, obtain the attitude information of each frame facial image to be measured, calculate average attitude information, be i.e. the attitude information of video sequence.

2. according to claim 1 a kind of based on image ratio to the head pose detection method with heading and attitude system, it is characterized in that, a ₀=0 °, b ₀=0 °, c ₀=0 °.

3. according to claim 1 and 2 a kind of based on image ratio to the head pose detection method with heading and attitude system, it is characterized in that, described face recognition technology is template matches or eigenface.