Method and apparatus for pattern recognition

1. A method for pattern recognition, comprising:

acquiring, by a processor, a two-dimensional image of a target object and acquiring, using a convolutional neural network model, a two-dimensional feature of the target object according to the two-dimensional image of the target object;

acquiring, by a depth camera, a three-dimensional image of the target object and acquiring a three-dimensional feature of the target object according to the three-dimensional image of the target object; and

identifying, by the processor, the target object according to the two-dimensional feature and the three-dimensional feature of the target object,

wherein:

the three-dimensional image of the target object comprises a three-dimensional curved surface of the target object;

the three-dimensional feature of the target object comprises at least one three-dimensional feature curve of the three-dimensional curved surface of the target, and

the at least one three-dimensional feature curve comprises:

a first curve where at least one plane or a part of the at least one plane intersects the three-dimensional curved surface of the target object or a set of points in the three-dimensional curved surface of the target object with a distance to the first curve less than or equal to a threshold; or

a second curve where at least one curved surface or a part of the at least one curved surface intersects the three-dimensional curved surface of the target object or a set of points in the three-dimensional curved surface of the target object with a distance to the second curve less than or equal to the threshold.

2. The method according to claim 1 ,

wherein the three-dimensional feature of the target object is concentrated in a first region, and

wherein the first region comprises a partial region of the three-dimensional curved surface of the target object, and the at least one three-dimensional feature curve is located in the first region.

3. The method according to claim 1 , wherein:

the at least one plane is perpendicular to a feature plane of the three-dimensional curved surface of the target object; and

the feature plane of the three-dimensional curved surface of the target object comprises a plane in which three feature points on the three-dimensional curved surface of the target object are located.

4. The method according to claim 3 , wherein:

the at least one plane intersects with the three-dimensional curved surface of the target object at a straight line; and

the straight line is perpendicular to the feature plane of the three-dimensional curved surface of the target object.

5. The method according to claim 3 , wherein, before acquiring the three-dimensional feature of the target object according to the three-dimensional image of the target object, the method further comprises

performing, under a three-dimensional coordinate system whose origin is a feature point on the three-dimensional curved surface of the target object, a pose adjustment to the three-dimensional curved surface of the target object,

wherein the pose adjustment comprises at least one of the following:

performing the pose adjustment according to a normal vector of the feature plane of the three-dimensional curved surface of the target object; and

selecting two feature points on the three-dimensional curved surface of the target object and performing the pose adjustment according to a vector where the two feature points are located.

6. The method according to claim 5 , wherein:

the performing the pose adjustment according to a normal vector of the feature plane of the three-dimensional curved surface of the target object comprises:

determining a rotation matrix R z for rotating the normal vector of the feature plane of the three-dimensional curved surface of the target object to a direction as same as a positive direction of a Z axis in the three-dimensional coordinate system, wherein an included angle between the normal vector and the positive direction of the Z axis is acute; and

rotating the three-dimensional curved surface of the target object according to the rotation matrix R z ; and

the performing the pose adjustment according to a vector where the two feature points are located comprises:

determining a rotation matrix R x for rotating the vector where the two feature points are located to a direction as same as a positive direction of an X axis in the three-dimensional coordinate system, wherein an included angle between the vector where the two feature points are located and the positive direction of the X axis is acute; and

rotating the three-dimensional curved surface of the target object according to the rotation matrix R x .

7. The method according to claim 1 , further comprising:

determining a first distance between the at least one three-dimensional feature curve of the target object and at least one three-dimensional feature curve corresponding to at least one three-dimensional image in a set of three-dimensional images of objects; and

determining a first weighted sum of the first distance, wherein the first weighted sum represents a three-dimensional image similarity between the three-dimensional image of the target object and the at least one three-dimensional image in the set of three-dimensional images of the objects.

8. The method according to claim 7 , wherein:

the acquiring a two-dimensional feature of the target object according to the two-dimensional image of the target object comprises acquiring the two-dimensional feature of the two-dimensional image of the target object and a two-dimensional feature of at least one two-dimensional image in a set of two-dimensional images of the objects; and

the method further comprises:

determining a second distance between the two-dimensional feature of the at least one two-dimensional image in the set of two-dimensional images for the objects and the two-dimensional feature of the two-dimensional image of the target object; and

determining a second weighted sum of the second distance, wherein the second weighted sum represents a two-dimensional similarity between the two-dimensional image of the target object and the at least one two-dimensional image in the set of two-dimensional images of the objects.

9. The method according to claim 8 , further comprising:

determining a third weighted sum of the two-dimensional similarity and the three-dimensional image similarity, wherein the third weighted sum represents a similarity between the target object and at least one object from the objects; and

identifying the target object in the objects according to the similarity.

10. The method according to claim 1 , wherein:

the target object comprises a target face;

the three-dimensional image of target object comprises a three-dimensional face curved surface of the target face;

the three-dimensional feature of the target object comprises at least one three-dimensional face feature curve of the target face;

the at least one three-dimensional face feature curve of the target face comprises:

a first curve where at least one plane or part of the at least one plane intersects the three-dimensional face curved surface of the target face or a set of points in the three-dimensional face curved surface of the target face with a distance to the first curve less than or equal to a threshold; or

a second curve where at least one curved surface or a part of the at least one curved surface intersects the three-dimensional face curved surface of the target face or a set of points in the three-dimensional face curved surface of the target face with a distance to the second curve less than or equal to the threshold.

11. The method according to claim 10 , wherein the three-dimensional face feature curve is located in upper half of the face comprising a tip of a nose.

12. The method according to claim 10 , wherein:

the at least one plane is perpendicular to a face feature plane of the target face; and

the face feature plane of the target face comprises a plane in which three of the following feature points are located:

one point or two points of an inner corner of a left eye, an outer corner of the left eye, an inner corner of a right eye, and an outer corner of the right eye;

at least one of a left corner of a mouth and a right corner of the mouth; and a tip of a nose.

13. The method according to claim 12 , wherein:

the at least one plane intersects the three-dimensional face curved surface of the target face at a straight line;

the straight line is perpendicular to the face feature plane of the target face; and

the straight line passes the tip of the nose on the target face.

14. The method according to claim 12 ,

wherein, before acquiring the three-dimensional feature of the target object according to the three-dimensional image of the target object, the method further comprises performing, under a three-dimensional coordinate system whose origin is a tip of a nose on a target face, a pose adjustment to the three-dimensional face curved surface of the target face,

wherein the pose adjustment comprises at least one of the following:

performing the pose adjustment according to a normal vector of the face feature plane of the target face; and

performing the pose adjustment according to a vector which is determined according to the left corner of the mouth and the right corner point of the mouth on the three-dimensional face curved surface of the target face.

15. The method according to claim 14 , wherein:

the performing the pose adjustment according to a normal vector of the face feature plane of the target face comprises:

determining a rotation matrix R z for rotating the normal vector of the face feature plane of the target face to a direction as same as a positive direction of a Z axis in the three-dimensional coordinate system, wherein an included angle between the normal vector and the positive direction of the Z axis is acute; and

rotating the three-dimensional face curved surface of the target face according to the rotation matrix R z ; and

the performing the pose adjustment according to a vector which is determined according to the left corner point of the mouth and the right corner point of the mouth on the target face comprises:

determining a rotation matrix R x for rotating the vector which is determined according to the left corner point of the mouth and the right corner point of the mouth on the target face to a direction as same as a positive direction of an X axis in the three-dimensional coordinate system, wherein an included angle between the determined vector and the positive direction of the X axis is acute; and

rotating the three-dimensional face curved surface of the target face according to the rotation matrix R x .

16. The method according to claim 10 , further comprising:

determining a first distance between the at least one three-dimensional face feature curve of the target face and the at least one three-dimensional face feature curve corresponding to at least one face in a set of faces; and

determining a first weighted sum of the first distance, wherein the first weighted sum represents a three-dimensional face similarity between the target face and the at least one face in the set of the faces.

17. The method according to claim 16 , wherein:

the acquiring the two-dimensional feature of the target object according to the two-dimensional image of the target object comprises acquiring a two-dimensional feature of the target face and a two-dimensional feature of at least one face in the faces using the convolutional neural network model;

the method further comprises:

determining a second distance between the two-dimensional feature of the at least one face in the set of faces and the two-dimensional feature of the target face; and

determining a second weighted sum of the second distance, wherein the second weighted sum represents a two-dimensional face similarity between the target face and at least one face in the set of faces.

18. The method according to claim 17 , further comprising:

determining a third weighted sum of the two-dimensional face similarity of the target face and the three-dimensional face similarity of the target face, wherein the third weighted sum represents a similarity between the target face and at least one face in a set of faces;

identifying the target face in the set of faces according to the similarity of the at least one face.

19. The method according to claim 6 , wherein the included angle is 10 degrees.

20. An apparatus for pattern recognition, comprising

an acquiring module and an identifying module,

wherein the acquiring module acquires a two-dimensional image of a target object using a convolutional neural network model, acquires a two-dimensional feature of the target object according to the two-dimensional image of the target object, acquires a three-dimensional image of the target object by a depth camera, and acquires a three-dimensional feature of the target object according to the three-dimensional image of the target object;

wherein the identifying module identifies the target object according to the two-dimensional feature and the three-dimensional feature of the target object obtained by the acquiring module, and

wherein:

the three-dimensional image of the target object comprises a three-dimensional curved surface of the target object;

the three-dimensional feature of the target object comprises at least one three-dimensional feature curve of the three-dimensional curved surface of the target object, and

the at least one three-dimensional feature curve comprises:

a first curve where at least one plane or a part of the at least one plane intersects the three-dimensional curved surface of the target object or a set of points in the three-dimensional curved surface of the target object with a distance to the first curve less than or equal to a threshold; or

a second curve where at least one curved surface or a part of the at least one curved surface intersects the three-dimensional curved surface of the target object or a set of points in the three-dimensional curved surface of the target object with a distance to the second curve less than or equal to the threshold.

21. A non-transitory computer readable storage medium that is stored with a computer program, wherein the computer program performs the method according to claim 1 when executed by the processor.

22. An apparatus for pattern recognition, comprising a memory, the processor of claim 1 , and a computer program stored in the memory and executable on the processor, wherein the computer program performs the method according to claim 1 when executed by the processor.