Living body detection method and system, computer-readable storage medium

1. A living body detection method, comprising:

acquiring a video including an object to be detected;

extracting at least two images to be detected from the video, and determining optical flow information according to the at least two images to be detected;

dividing each image to be detected into a foreground image and a background image according to the optical flow information;

using a classifier to perform category judgment on the foreground image and the background image to obtain a category distribution of the foreground image and a category distribution of the background image, wherein the category distribution of the foreground image indicates a probability that the foreground image belongs to an attack category and a probability that the foreground image belongs to a living-body category, and the category distribution of the background image indicates a probability that the background image belongs to the attack category and a probability that the background image belongs to the living-body category;

respectively acquiring a first probability that the foreground image belongs to the living-body category and a second probability that the background image belongs to the living-body category according to the category distribution of the foreground image and the category distribution of the background image; and

fusing the first probability and the second probability through a preset fusion algorithm to obtain a probability that the object to be detected is a living body.

2. The method according to claim 1 , wherein the classifier comprises a neural network model; and

using the classifier to perform category judgment on the foreground image and the background image to obtain the category distribution of the foreground image and the category distribution of the background image comprises:

inputting the foreground image and the background image into the neural network model; and

using the neural network model to perform the category judgment on the foreground image and the background image to obtain the category distribution of the foreground image and the category distribution of the background image.

3. The method according to claim 2 , further comprising:

acquiring sample video data, wherein the sample video data comprises foreground sets and background sets corresponding to sample videos, and the sample videos comprise a living-body video and an attack video;

dividing the sample video data into a training set, a validation set and a test set;

training a 3D neural network model by using the training set;

stopping training when the 3D neural network model is adopted to classify the validation set to obtain a classification accuracy and the classification accuracy is greater than a preset accuracy threshold; and

testing the 3D neural network model by using the test set, and determining the 3D neural network model obtained after the testing as the neural network model used for living body detection.

4. The method according to claim 3 , wherein after acquiring the sample video data, the method further comprises:

performing a pre-processing operation on sample images extracted from the sample videos,

wherein the pre-processing operation at least comprises one of following operations:

flipping, cutting or resolution adjustment.

5. The method according to claim 1 , wherein dividing each image to be detected into the foreground image and the background image according to the optical flow information comprises:

determining that each pixel belongs to foreground or background according to a motion rate of each pixel in the optical flow information; and

combining pixels belonging to the foreground in each image to be detected into the foreground image, and combining pixels belonging to the background in each image to be detected into the background image.

6. The method according to claim 5 , wherein determining that each pixel belongs to the foreground or the background according to the motion rate of each pixel in the optical flow information comprises:

determining the motion rate of each pixel in each image to be detected according to the optical flow information; and

judging whether a motion rate of a pixel is greater than a preset rate threshold, determining that the pixel belongs to the foreground in a case where the motion rate of the pixel is greater than the preset rate threshold, and determining that the pixel belongs to the background in a case where the motion rate of the pixel is not greater than the preset rate threshold.

7. The method according to claim 6 , wherein determining the motion rate of each pixel in each image to be detected according to the optical flow information comprises:

acquiring X-channel information and Y-channel information of each pixel in each image to be detected, wherein the X-channel information and the Y-channel information respectively represent a motion rate in an X direction and a motion rate in a Y direction; and

determining a larger value between the motion rate of the X direction and the motion rate of the Y direction as the motion rate of each pixel.

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

judging whether the probability that the object to be detected is the living body is greater than a preset living-body probability threshold; and

determining that the object to be detected is the living body in a case where the probability that the object to be detected is the living body is greater than the preset living-body probability threshold.

9. A living body detection system, comprising:

a video capture device;

a processor; and

a storage device,

wherein the video capture device is configured to capture a video including an object to be detected;

wherein computer programs are stored in the storage device that, when executed by the processor, cause the processor to:

extract at least two images to be detected from the video, and determine optical flow information according to the at least two images to be detected,

divide each image to be detected into a foreground image and a background image according to the optical flow information,

use a classifier to perform category judgment on the foreground image and the background image to obtain a category distribution of the foreground image and a category distribution of the background image, wherein the category distribution of the foreground image indicates a probability that the foreground image belongs to an attack category and a probability that the foreground image belongs to a living-body category, and the category distribution of the background image indicates a probability that the background image belongs to the attack category and a probability that the background image belongs to the living-body category,

respectively acquire a first probability that the foreground image belongs to the living-body category and a second probability that the background image belongs to the living-body category according to the category distribution of the foreground image and the category distribution of the background image, and

fuse the first probability and the second probability through a preset fusion algorithm to obtain a probability that the object to be detected is a living body.

10. A non-transitory computer-readable storage medium, wherein computer programs are stored on the non-transitory computer-readable storage medium; that, when executed by a processor, perform operations comprising:

acquiring a video including an object to be detected;

extracting at least two images to be detected from the video, and determining optical flow information according to the at least two images to be detected;

dividing each image to be detected into a foreground image and a background image according to the optical flow information;

using a classifier to perform category judgment on the foreground image and the background image to obtain a category distribution of the foreground image and a category distribution of the background image, wherein the category distribution of the foreground image indicates a probability that the foreground image belongs to an attack category and a probability that the foreground image belongs to a living-body category, and the category distribution of the background image indicates a probability that the background image belongs to the attack category and a probability that the background image belongs to the living-body category;

respectively acquiring a first probability that the foreground image belongs to the living-body category and a second probability that the background image belongs to the living-body category according to the category distribution of the foreground image and the category distribution of the background image; and

fusing the first probability and the second probability through a preset fusion algorithm to obtain a probability that the object to be detected is a living body.