METHODS AND APPARATUSES FOR DETERMINING BOUNDING BOX OF TARGET OBJECT, MEDIA, AND DEVICES

• US 20200134859A1
• Filed: 12/31/2019
• Published: 04/30/2020
• Est. Priority Date: 11/21/2017
• Status: Active Grant

• Current Assignee: Beijing SenseTime Technology Development Co., Ltd.
• Original Assignee: Beijing SenseTime Technology Development Co., Ltd.
• Inventors: Li, Buyu, Li, Quanquan, Yan, Junjie
• Granted Patent: US 11,348,275 B2
• Time in Patent Office: Days
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/048 Activation functions
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/30196 Human being; Person
• CPC Class Codes: G06T 2210/12 Bounding box
• CPC Class Codes: G06T 7/73 using feature-based methods
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

Embodiments of the present application disclose methods and apparatuses for determining a bounding box of a target object, media, and devices. The method includes: obtaining attribute information of each of a plurality of key points of a target object; and determining a bounding box position of the target object according to the attribute information of each of the plurality of key points of the target object and to a preset neural network. The implementations of the present application can improve the efficiency and accuracy of determining a bounding box of a target object.

Claims

1. A method for determining a bounding box of a target object, comprising:

obtaining attribute information of each of a plurality of key points of a target object; and

determining a bounding box position of the target object according to the attribute information of each of the plurality of key points of the target object and to a preset neural network.

2. The method according to claim 1, wherein the attribute information of the key point comprises coordinate information and a presence determination value.

3. The method according to claim 2, wherein determining the bounding box position of the target object according to the attribute information of each of the plurality of key points of the target object and to the preset neural network comprises:

determining at least one valid key point from the plurality of key points according to the attribute information of each of the plurality of key points;

processing, according to the attribute information of each of the at least one valid key point, the attribute information of the plurality of key points to obtain processed attribute information of the plurality of key points; and

inputting the processed attribute information of the plurality of key points to the preset neural network for processing to obtain the bounding box position of the target object.

4. The method according to claim 3, wherein the processed attribute information of the plurality of key points comprises:

processed attribute information of each of the at least one valid key point, and attribute information of key points other than the at least one valid key point in the plurality of key points.

5. The method according to claim 3, wherein processing, according to the attribute information of each of the at least one valid key point, the attribute information of the plurality of key points to obtain processed attribute information of the plurality of key points comprises:

determining a reference coordinate according to coordinate information comprised in the attribute information of each of the at least one valid key point; and

determining coordinate information in the processed attribute information of each valid key point according to the reference coordinate and to the coordinate information in the attribute information of each of the at least one valid key point.

6. The method according to claim 5, wherein determining the reference coordinate according to coordinate information comprised in the attribute information of each of the at least one valid key point comprises:

performing averaging processing on coordinates corresponding to the coordinate information of each of the at least one valid key point to obtain the reference coordinate, and/or

determining the coordinate information in the processed attribute information of each valid key point according to the reference coordinate and to the coordinate information in the attribute information of each of the at least one valid key point comprises;

determining, with the reference coordinate as an origin point, processed coordinate information corresponding to the coordinate information of each of the at least one valid key point.

7. The method according to claim 5, wherein inputting the processed attribute information of the plurality of key points to the preset neural network for processing to obtain the bounding box position of the target object comprises:

inputting the processed attribute information of the plurality of key points to the preset neural network for processing to obtain output position information; and

determining the bounding box position of the target object according to the reference coordinate and the output position information.

8. The method according to claim 1, wherein the neural network comprises at least two full connection layers.

9. The method according to claim 1, wherein the neural network comprises three full connection layers, wherein an activation function of at least one of the first full connection layer and the second full connection layer of the three full connection layers comprises a Rectified Linear Unit (ReLu) activation function.

10. A computer-readable storage medium having instructions stored thereon, wherein when the instructions are executed by a processor, the method according to claim 1 is executed.

11. An electronic device comprising:

a processor; and

a computer-readable storage medium having stored thereon instructions that, when executed by the processor, cause the processor to;

obtain attribute information of each of a plurality of key points of a target object; and

determine a bounding box position of the target object according to the attribute information of each of the plurality of key points of the target object obtained by the obtaining module and a preset neural network.

12. The electronic device according to claim 11, wherein the attribute information of the key point comprises coordinate information and a presence determination value.

13. The electronic device according to claim 12, wherein determining the bounding box position of the target object according to the attribute information of each of the plurality of key points of the target object and to the preset neural network comprises:

determining at least one valid key point from the plurality of key points according to the attribute information of each of the plurality of key points;

processing, according to the attribute information of each of the at least one valid key point, the attribute information of the plurality of key points to obtain processed attribute information of the plurality of key points; and

inputting the processed attribute information of the plurality of key points to the preset neural network for processing to obtain the bounding box position of the target object,

wherein the processed attribute information of the plurality of key points comprises processed attribute information of each of the at least one valid key point and attribute information of key points other than the at least one valid key point in the plurality of key points.

14. The electronic device according to claim 13, wherein processing, according to the attribute information of each of the at least one valid key point, the attribute information of the plurality of key points to obtain processed attribute information of the plurality of key points comprises:

determining a reference coordinate according to coordinate information comprised in the attribute information of each of the at least one valid key point; and

determining coordinate information in the processed attribute information of each valid key point according to the reference coordinate and to the coordinate information in the attribute information of each of the at least one valid key point.

15. The electronic device according to claim 14, wherein determining the reference coordinate according to coordinate information comprised in the attribute information of each of the at least one valid key point comprises:

performing averaging processing on coordinates corresponding to the coordinate information of each of the at least one valid key point to obtain the reference coordinate, and/or

determining the coordinate information in the processed attribute information of each valid key point according to the reference coordinate and to the coordinate information in the attribute information of each of the at least one valid key point comprises;

determining, with the reference coordinate as an origin point, processed coordinate information corresponding to the coordinate information of each of the at least one valid key point.

16. The electronic device according to claim 14, wherein inputting the processed attribute information of the plurality of key points to the preset neural network for processing to obtain the bounding box position of the target object comprises:

inputting the processed attribute information of the plurality of key points to the preset neural network for processing to obtain output position information; and

determining the bounding box position of the target object according to the reference coordinate and the output position information.

17. The electronic device according to claim 11, wherein the neural network comprises at least two full connection layers.

18. The electronic device according to claim 11, wherein the neural network comprises three full connection layers, wherein an activation function of at least one of the first full connection layer and the second full connection layer of the three full connection layers comprises a Rectified Linear Unit (ReLu) activation function.