Granted Patent B2

US 12,430,930 · App. 17/720,024 · Granted Sep 30, 2025

Inventors: Yuqian Zeng (Shanghai, CN); Tao Zhang (Shanghai, CN)

Assignee: Black Sesame Technologies Inc.

G06V20/597G06V10/26G06V10/764G06V10/7715G06V10/82G06V40/161

Abstract

Claims (57)

1. A safety belt detection method, comprising:

obtaining an image to be detected, further comprising:

capturing an original image by a camera;

detecting a face in the original image based on a predetermined algorithm to obtain a face bounding box and a head posture information;

determining whether a part below the face in the original image meets a predetermined size;

determining a detection result as insufficient detection area if the predetermined size is not met;

dynamically cropping the part below the face in the original image based on the face bounding box and the head posture information to obtain a cropped image if the predetermined size is met; and

preprocessing the cropped image to obtain the image to be detected;

inputting the image to be detected into a detection network, wherein the detection network comprises an image classification branch network and an image segmentation branch network;

obtaining a classification result output from the image classification branch network and a segmentation image output from the image segmentation branch network, wherein the classification result indicates whether a driver is wearing a safety belt, and the segmentation image indicates a position information of the safety belt;

obtaining a detection result of the safety belt based on the classification result and the segmentation image; wherein the detection result indicates whether the driver wears the safety belt normatively; and

the segmentation image is reduced by a predetermined proportion relative to the image to be detected without encoding as the image to be detected input to the detection network, and pixel positions of the segmentation image are mapped corresponding to pixel positions of the image to be detected, to locate the positions of seat belt pixels in the image to be detected.

2. The safety belt detection method of claim 1 , wherein after obtaining an original image through a camera, the safety belt detection method further comprises: detecting a human hand in the original image based on a human hand detection algorithm to obtain a human hand bounding box: obtaining a safety belt bounding box in the original image based on a position and a size of the face bounding box; and determining that the detection result is that the safety belt is covered when the hand bounding box is located at a predetermined position relative to the face bounding box and an overlapped area of the hand bounding box and the safety belt bounding box is greater than a set threshold.

3. The safety belt detection method of claim 1 , wherein the detection network further comprises a backbone network; and

before obtaining a classification result output from the image classification branch network and obtaining a segmentation image output from the image segmentation branch network, the method further comprises:

obtaining a convolution of the image to be detected by the backbone network to generate a feature map of the image to be detected, and using the feature map as an input of the image classification branch network and the image segmentation branch network.

4. The safety belt detection method of claim 1 , wherein the step of obtaining a detection result of the safety belt based on the classification result and the segmentation image further comprises:

fitting a centerline of the safety belt from the segmentation image according to a predetermined rule;

calculating parameters related to the centerline; and

obtaining the detection result of the safety belt based on the parameters and the classification result.

5. The safety belt detection method of claim 4 , wherein the step of fitting a centerline of the safety belt from the segmentation image according to a predetermined rule further comprises:

obtaining a center point of the safety belt area in a row or column either row by row or column by column according to the segmentation image;

fitting all center points in the segmentation image to obtain the centerline of the safety belt;

wherein the parameters related to the centerline comprises: an included angle between the centerline and a predetermined direction, and a relative position between the centerline and the driver's shoulder.

6. The safety belt detection method of claim 1 , wherein before obtaining the image to be detected, the method further comprises:

selecting training images, and training the image classification branch network, wherein the training images are divided into label images of wearing safety belt and label images of not wearing safety belt;

reducing the label images of wearing safety belt according to the predetermined proportion, and using the reduced label images to train the image segmentation branch network.

7. A safety belt detection apparatus, comprising a processor, wherein the processor performs the steps of:

obtaining an image to be detected, further comprising:

capturing an original image by a camera;

detecting a face in the original image based on a predetermined algorithm to obtain a face bounding box and a head posture information;

determining whether a part below the face in the original image meets a predetermined size;

determining a detection result as insufficient detection area if the predetermined size is not met;

dynamically cropping the part below the face in the original image based on the face bounding box and the head posture information to obtain a cropped image if the predetermined size is met; and

preprocessing the cropped image to obtain the image to be detected;

inputting the image to be detected into a detection network, wherein the detection network comprises an image classification branch network and an image segmentation branch network;

obtaining a classification result output from the image classification branch network and a segmentation image output from the image segmentation branch network, wherein the classification result indicates whether a driver is wearing a safety belt, and the segmentation image indicates a position information of the safety belt;

obtaining a detection result of the safety belt based on the classification result and the segmentation image: wherein the detection result indicates whether the driver wears the safety belt normatively; and

the segmentation image is reduced by a predetermined proportion relative to the image to be detected without encoding as the image to be detected input to the detection network, and pixel positions of the segmentation image are mapped corresponding to pixel positions of the image to be detected, to locate the positions of seat belt pixels in the image to be detected.

8. The safety belt detection apparatus of claim 7 , wherein the processor further performs the steps of detecting a human hand in the original image based on a human hand detection algorithm to obtain a human hand bounding box;

obtaining a safety belt bounding box in the original image based on a position and a size of the face bounding box; and determining that the detection result is that the safety belt is covered when the hand bounding box is located at a predetermined position relative to the face bounding box and an overlapping area of the hand bounding box and the safety belt hounding box is greater than a set threshold.

9. The safety belt detection apparatus of claim 7 , wherein the detection network further comprises a backbone network; and

the image detection unit further comprises:

a feature extraction unit, configured to obtain a convolution of the image to be detected by the backbone network to generate a feature map of the image to be detected, and use the feature map as an input of the image classification branch network and the image segmentation branch network.

10. The safety belt detection apparatus of claim 7 , wherein the processor further performs the steps of:

fitting a centerline of the safety belt from the segmentation image according to a predetermined rule; and

calculating parameters related to the centerline; and

obtaining the detection result of the safety belt based on the parameters and the classification result.

11. The safety belt detection apparatus of claim 10 , wherein the processor further performs the steps of:

obtaining a center point of the safety belt area in a row or column, row by row, or column according to the segmentation image; and

fitting all center points in the segmentation image to obtain the centerline of the safety belt;

wherein the parameters related to the centerline comprises: an included angle between the centerline and a predetermined direction, and a relative position between the centerline and the driver's shoulder.

12. The safety belt detection apparatus of claim 7 , wherein the processor further performs the steps of:

selecting training images, and train the image classification branch network, wherein the training images are divided into label images of wearing safety belt and label images of not wearing safety belt; and

reducing the label images of wearing safety belt according to the predetermined proportion, and use the reduced label images to train the image segmentation branch network.

13. A computer device, comprising a memory on which a computer program is stored and a processor, wherein the processor, when executing the computer program, implements the safety belt detection method of claim 1 .

14. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the safety belt detection method of claim 1 .

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (42)