Human body identification method, electronic device and storage medium

Patent metadata

Field	Value
Publication number	US-11854237-B2
Application number	US-202117353324-A
Country	US
Kind code	B2
Filing date	Jun 21, 2021
Priority date	Dec 11, 2020
Publication date	Dec 26, 2023
Grant date	Dec 26, 2023

Abstract

A human body identification method, an electronic device and a storage medium, related to the technical field of artificial intelligence such as computer vision and deep learning, are provided. The method includes: inputting an image to be identified into a human body detection model, to obtain a plurality of preselected detection boxes; identifying a plurality of key points from each of the preselected detection boxes respectively according to a human body key point detection model, and obtaining a key point score of each of the key points; determining a target detection box from each of the preselected detection boxes, according to a number of the key points whose key point scores meet a key point threshold; and inputting the target detection box into a human body key point classification model, to obtain a human body identification result for the image to be identified.

First claim

What is claimed is:

1. A human body identification method, comprising:
inputting an image to be identified into a human body detection model, to obtain a plurality of preselected detection boxes;
identifying a plurality of key points from each of the preselected detection boxes respectively according to a human body key point detection model, and obtaining a key point score of each of the key points;
determining a target detection box from each of the preselected detection boxes, according to a number of the key points whose key point scores meet a key point threshold; and
inputting the target detection box into a human body key point classification model, to obtain a human body identification result for the image to be identified,
wherein the key point threshold comprises a low keypoint threshold, and the determining the target detection box from each of the preselected detection boxes, according to the number of the key points whose key point scores meet the key point threshold, comprises:
determining a corresponding preselected detection box as the target detection box, in a case where a first key point number is larger than a preset number, wherein the first key point number is a number of the key points whose key point scores are larger than a low key point threshold,
wherein the key point threshold further comprises a high key point threshold, and the method further comprises:
determining the corresponding preselected detection box as a second high confidence detection box, in a case where the first key point number is greater than zero and less than or equal to the preset number, and a second key point number is greater than or equal to a preset proportion of the first key point number, wherein the second key point number is a number of the key points whose key point scores are lager than the high key point threshold; and
extracting a second human body image area from the image to be identified, according to a size and a position of the second high confidence detection box.

2. The method of claim 1 , wherein the inputting the image to be identified into the human body detection model, to obtain the plurality of preselected detection boxes, comprises:
inputting the image to be identified into the human body detection model, to obtain a plurality of initial detection boxes and a detection box score of each of the initial detection boxes; and
determining a corresponding initial detection box as the preselected detection box, in a case where the detection box score is greater than or equal to a low detection box threshold and less than or equal to a high detection box threshold.

3. The method of claim 2 , further comprising:
determining the corresponding initial detection box as a first high confidence detection box, in a case where the detection box score is greater than the high detection box threshold; and
extracting a first human body image area from the image to be identified, according to a size and a position of the first high confidence detection box.

4. The method of claim 1 , wherein the target detection box comprises a plurality of target detection boxes, and the inputting the target detection box into the human body key point classification model, to obtain the human body identification result for the image to be identified, comprises:
performing key-point-based non maximum suppression (NMS) processing on each of the target detection boxes; and
inputting the target detection box after the NMS processing into the human body key point classification model, to obtain the human body identification result.

5. A method for generating a human body key point classification model, comprising:
acquiring a positive sample image, wherein the positive sample image corresponds to at least one human body labeling box comprising a key point label;
generating a negative sample image based on the positive sample image;
inputting the negative sample image into a human body key point detection model, to obtain a negative sample labeling box; and
training a machine learning model according to the human body labeling box and the negative sample labeling box, to generate a human body key point classification model, wherein the human body key point classification model is used for the human body identification method of claim 1 .

6. The method of claim 5 , wherein the generating the negative sample image based on the positive sample image, comprises:
generating a sample detection box randomly on the positive sample image; and
taking a negative sample human body image from the positive sample image according to a size and a position of the sample detection box, in a case where a distance between the sample detection box and the respective human body labeling boxes is smaller than a preset distance.

7. An electronic device, comprising:
at least one processor; and
a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform operations of:
inputting an image to be identified into a human body detection model, to obtain a plurality of preselected detection boxes;
identifying a plurality of key points from each of the preselected detection boxes respectively according to a human body key point detection model, and obtaining a key point score of each of the key points;
determining a target detection box from each of the preselected detection boxes, according to a number of the key points whose key point scores meet a key point threshold; and
inputting the target detection box into a human body key point classification model, to obtain a human body identification result for the image to be identified,
wherein the key point threshold comprises a low key point threshold, and the determining the tarot detection box from each of the preselected detection boxes, according to the number of the key points whose key point scores meet the key point threshold, comprises:
determining a corresponding preselected detection box as the target detection box, in a case where a first key point number is larger than a preset number, wherein the first key point number is a number of the key points whose key point scores are larger than a low key point threshold,
wherein the key point threshold further comprises a high key point threshold, and the instructions are executable by the at least one processor to enable the at least one processor to further perform operations of:
determining the corresponding preselected detection box as a second high confidence detection box, in a case where the first key point number is greater than zero and less than or equal to the preset number, and a second key point number is greater than or equal to a preset proportion of the first key point number, wherein the second key point number is a number of the key points whose key point scores are larger than the high key point threshold; and
extracting a second human body image area from the image to be identified, according to a size and a position of the second high confidence detection box.

8. The electronic device of claim 7 , wherein the inputting the image to be identified into the human body detection model, to obtain the plurality of preselected detection boxes, comprises:
inputting the image to be identified into the human body detection model, to obtain a plurality of initial detection boxes and a detection box score of each of the initial detection boxes; and
determining a corresponding initial detection box as the preselected detection box, in a case where the detection box score is greater than or equal to a low detection box threshold and less than or equal to a high det

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Lu Zipeng
• Wang Jian
• Yuan Yuchen
• Sun Hao
• Ding Errui

Classifications

• G06V10/25 Primary

  Determination of region of interest [ROI] or a volume of interest [VOI] · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V10/757

  Matching configurations of points or features · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title