Safety helmet correct wearing detection method, system and device for building construction site and storage medium

Abstract

The invention discloses a helmet correct wearing detection method, system and device for a building construction site and a storage medium, and is applied to the field of building construction, and the method comprises the steps: carrying out the reflection removal of a to-be-detected image based on the gray comparison and preprocessing of RGB three channels; obtaining a to-be-recognized face image in the to-be-detected image after the light reflection removal processing, and matching the identity of the to-be-recognized face image and a corresponding safety helmet wearing standard based on a multi-task regression face and posture feature recognition model; inputting a head image corresponding to the to-be-recognized face image with successful identity matching into a safety helmet wearing detection model based on YOLOv4, and outputting a safety helmet wearing detection result; and based on the safety helmet wearing detection result, judging whether the safety helmet wearing of the face image to be recognized is consistent with the safety helmet wearing standard corresponding to the matched identity by utilizing RGB three-channel pixel value comparison, and outputting a correct safety helmet wearing detection result. According to the invention, the safety helmet wearing detection precision is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of building construction, and in particular to a method, system, equipment and storage medium for detecting the correct wearing of a safety helmet at a building construction site. Background Art

[0002] The correct wearing of safety helmets at construction sites has always been a core requirement in the construction field. The correct wearing of safety helmets not only makes the division of labor of construction site workers clear at a glance, but also greatly protects the lives of different types of workers. However, most grassroots personnel at construction sites lack this awareness, resulting in a complicated division of labor at construction sites. What's more, if they do not wear safety helmets, their own safety cannot be effectively guaranteed. In addition, construction site supervisors cannot provide 24-hour and all-round monitoring and reminders. Therefore, how to achieve automatic detection of safety helmet wearing has always been a hot topic.

[0003] The existing automatic detection method for wearing hard hats generally uses a camera to obtain the on-site situation of the construction site in real time, and uses image recognition technology to determine whether the workers are wearing hard hats. This technology has the following defects: there are many reflective materials on the construction site, which can easily cause misjudgment caused by image reflection; it is impossible to effectively judge the workers of the construction unit, and other people or objects that do not belong to the construction unit may be detected; the features extracted in the hard hat recognition process are single, and the recognition accuracy needs to be improved; it can only identify whether a hard hat is worn, but cannot identify whether the hard hats worn by different types of workers meet the wearing standards of their identities.

[0004] To this end, how to provide a method, system, equipment and storage medium for detecting the correct wearing of safety helmets for construction sites that can effectively solve the problems in existing safety helmet wearing detection technology that are affected by image reflection and single feature extraction, as well as the inability to effectively judge the staff of the construction unit and the inability to identify whether the safety helmets worn by different types of staff meet the wearing standards of their identities is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention proposes a method, system, device and storage medium for detecting the correct wearing of a safety helmet at a construction site.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for detecting whether a safety helmet is worn correctly at a construction site, comprising:

[0008] Step 1: Based on the grayscale contrast and preprocessing of the RGB three channels, the image to be tested is de-reflected;

[0009] Step 2: Obtain the face image to be identified in the image to be tested after de-reflection processing, and match the identity of the face image to be identified and the corresponding helmet wearing standard based on the face and posture feature recognition model of multi-task regression;

[0010] Step 3: Input the head image corresponding to the face image to be identified with successful identity matching into the helmet wearing detection model based on YOLOv4, and output the helmet wearing detection result;

[0011] Step 4: Based on the helmet wearing detection results, use the RGB three-channel pixel value comparison to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, and output the helmet correct wearing detection result.

[0012] Optionally, in step 1, based on grayscale contrast and preprocessing of the RGB three channels, de-reflection processing is performed on the image to be tested, specifically:

[0013] Decompose the image to be tested into three channels of RGB, and obtain grayscale images of the three channels of RGB respectively;

[0014] Compare the grayscale values ​​of the RGB three-channel grayscale images under the same pixel, combine the pixel points of the channel grayscale image with the smallest grayscale value under each pixel, and obtain the reflective image to be calibrated;

[0015] The grayscale value of each pixel in the reflective image to be calibrated is compared with the preset reflective grayscale threshold, the grayscale value of the pixel greater than the preset reflective grayscale threshold is set to 70 and calibrated as white, and the grayscale value of the pixel less than or equal to the preset reflective grayscale threshold is set to 0 and calibrated as black, to obtain a filtered reflective image;

[0016] Subtract each pixel in the image to be tested from that in the filtered reflection image to obtain the image to be tested after de-reflection processing.

[0017] Optionally, in step 2, before the face and posture feature recognition model based on multi-task regression matches the identity of the face image to be identified and the corresponding helmet wearing standard, it also includes: a segmentation feature extraction network based on the image segmentation model and an image classification network to obtain face pixels and sub-face pixels to determine the face area.

[0018] Optionally, in step 2, the face and posture feature recognition model based on multi-task regression includes: a face feature recognition network, a face posture feature recognition network, a distance posture recognition network, and an angle posture recognition network;

[0019] Inputting the face image to be recognized into the face feature recognition network and the posture feature recognition network respectively to obtain face feature information and posture feature information;

[0020] The posture feature information is respectively input into the distance posture recognition network and the angle posture recognition network to obtain the distance posture information and the angle posture information under the posture feature information, which are combined with the facial feature information to obtain the facial feature information corresponding to the distance posture information and the angle posture information under the posture feature information, so as to perform identity matching in the facial feature database.

[0021] Optionally, in step 3, the helmet wearing detection model based on YOLOv4 is specifically:

[0022] The SPPNet network is used to extract contextual features of head images using multiple maximum poolings at different scales.

[0023] The PANet network is used to perform parameter fusion of multiple feature maps of different scales output by the SPPNet network through convolution, up- and down-sampling, feature fusion, and stacking.

[0024] Helmet wearing detection is performed based on the features after parameter fusion.

[0025] Optionally, the pooling kernels of the maximum pooling include: 13x13, 9x9, 5x5, 1x1.

[0026] Optionally, in step 4, based on the helmet wearing detection result, the RGB three-channel pixel value comparison is used to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, specifically:

[0027] The feature extraction network is used to output the helmet wearing detection result as the head image with the helmet worn to perform helmet cropping;

[0028] The RGB three-channel pixel value at the center of the cropped helmet image is obtained, and the difference is made with the RGB three-channel pixel value of the helmet wearing standard corresponding to the matched identity of the face image to be identified. If the difference is within the preset error threshold, it is judged as consistent; if the difference is not within the preset error threshold, it is judged as inconsistent.

[0029] The present invention also provides a safety helmet correct wearing detection system for a construction site using a safety helmet correct wearing detection method for a construction site, comprising:

[0030] De-reflection module: used for grayscale contrast and preprocessing based on RGB three channels to perform de-reflection processing on the image to be tested;

[0031] Identity matching and helmet wearing standard recognition module: obtain the face image to be recognized in the image to be tested after de-reflection processing, and match the identity of the face image to be recognized and the corresponding helmet wearing standard based on the face and posture feature recognition model of multi-task regression;

[0032] Helmet wearing detection module: The head image corresponding to the face image to be identified with successful identity matching is input into the helmet wearing detection model based on YOLOv4, and the helmet wearing detection result is output;

[0033] Helmet correct wearing detection module: Based on the helmet wearing detection results, the RGB three-channel pixel value comparison is used to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, and the helmet correct wearing detection results are output.

[0034] The present invention also provides an electronic device, comprising:

[0035] Memory for storing computer programs;

[0036] The processor is used to implement the steps of a method for detecting the correct wearing of a safety helmet at a construction site when executing a computer program.

[0037] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for detecting the correct wearing of a safety helmet at a construction site are implemented.

[0038] Through the above technical solutions, it can be known that compared with the prior art, the present invention proposes a method, system, device and storage medium for detecting the correct wearing of safety helmets for construction sites. Through grayscale contrast and preprocessing based on RGB three channels, the problem of misjudgment caused by image reflection due to the large number of reflective materials in the construction site is solved; through the face and posture feature recognition model based on multi-task regression, the identity of the face image to be identified and the corresponding safety helmet wearing standard are matched, and the problem that the prior art cannot effectively judge the staff of the construction unit and may detect other personnel or objects that do not belong to the construction unit is solved. On the basis of further improving the accuracy of face recognition; through the safety helmet wearing detection model based on YOLOv4, multiple contextual features of maximum pooling of different scales are extracted and fused, which effectively improves the accuracy of safety helmet wearing recognition; by using the RGB three-channel pixel value comparison, it is judged whether the safety helmet wearing of the face image to be identified is consistent with the safety helmet wearing standard corresponding to the matched identity, which solves the problem of not being able to identify whether the safety helmets worn by different types of staff meet the wearing standards of their identities. And through the above-mentioned technical features, the present invention conducts three gradual inspections, namely, the inspection of non-construction unit staff, the inspection of safety helmets not being worn, and the inspection of safety helmets not being worn correctly, which further improves the accuracy of safety helmet inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Embodiment 1:

[0043] Embodiment 1 of the present invention discloses a method for detecting the correct wearing of a safety helmet at a construction site. Figure 1 As shown, including:

[0044] Step 1: Based on the grayscale contrast and preprocessing of the RGB three channels, the image to be tested is de-reflected.

[0045] Based on the grayscale contrast and preprocessing of the RGB three channels, the image to be tested is de-reflected, specifically:

[0046] Decompose the image to be tested into three channels of RGB, and obtain grayscale images of the three channels of RGB respectively;

[0047] Compare the grayscale values ​​of the RGB three-channel grayscale images under the same pixel, combine the pixel points of the channel grayscale image with the smallest grayscale value under each pixel, and obtain the reflective image to be calibrated;

[0048] Compare the grayscale value of each pixel in the reflective image to be calibrated with the preset reflective grayscale threshold (taken as 210), set the grayscale value of the pixel greater than the preset reflective grayscale threshold to 70, calibrate it as white (reflective part), set the grayscale value of the pixel less than or equal to the preset reflective grayscale threshold to 0, calibrate it as black (non-reflective part), and obtain the filtered reflective image;

[0049] Subtract each pixel in the image to be tested from that in the filtered reflection image to obtain the image to be tested after de-reflection processing.

[0050] In order to prevent the color of the reflective part from being too dark, which is not conducive to subsequent de-reflection processing, after calibrating the reflective part and the non-reflective part, it also includes: expanding all pixels of the calibrated image by 0.5-0.7 times to obtain a filtered reflective image.

[0051] Step 2: Get the face image to be identified in the image to be tested after de-reflection processing, and match the identity of the face image to be identified and the corresponding helmet wearing standard based on the face and posture feature recognition model of multi-task regression. (At this point, the first detection is completed, and the situation of non-construction unit personnel entering is reported)

[0052] Before the face and posture feature recognition model based on multi-task regression matches the identity of the face image to be recognized and the corresponding helmet wearing standard, it also includes: a segmentation feature extraction network based on the image segmentation model and an image classification network to obtain face pixels and sub-face pixels to determine the face area.

[0053] Face and posture feature recognition model based on multi-task regression, including: face feature recognition network, face posture feature recognition network, distance posture recognition network and angle posture recognition network;

[0054] Inputting the face image to be recognized into the face feature recognition network and the posture feature recognition network respectively to obtain face feature information and posture feature information;

[0055] The posture feature information is respectively input into the distance posture recognition network and the angle posture recognition network to obtain the distance posture information and the angle posture information under the posture feature information, which are combined with the facial feature information to obtain the facial feature information corresponding to the distance posture information and the angle posture information under the posture feature information, so as to perform identity matching in the facial feature database.

[0056] Step 3: Input the head image corresponding to the face image to be identified that has successfully matched the identity into the helmet wearing detection model based on YOLOv4, and output the helmet wearing detection result. (At this point, the second detection is completed, and the situation that XXX is not wearing a helmet is reported)

[0057] The helmet wearing detection model based on YOLOv4 is as follows:

[0058] The SPPNet network is used to extract contextual features of head images using multiple maximum poolings at different scales.

[0059] The PANet network is used to perform parameter fusion of multiple feature maps of different scales output by the SPPNet network through convolution, up- and down-sampling, feature fusion, and stacking.

[0060] Helmet wearing detection is performed based on the features after parameter fusion.

[0061] The pooling kernels of maximum pooling include: 13x13, 9x9, 5x5, 1x1.

[0062] Step 4: Based on the helmet wearing detection results, use the RGB three-channel pixel value comparison to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, and output the helmet correct wearing detection result. (At this point, the third test is completed, and the situation that XXX does not wear the helmet correctly is reported)

[0063] Based on the helmet wearing detection results, the RGB three-channel pixel value comparison is used to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, specifically:

[0064] The feature extraction network is used to output the helmet wearing detection result as the head image with the helmet worn to perform helmet cropping;

[0065] The RGB three-channel pixel value at the center of the cropped helmet image is obtained, and the difference is made with the RGB three-channel pixel value of the helmet wearing standard corresponding to the matched identity of the face image to be identified. If the difference is within the preset error threshold, it is judged as consistent; if the difference is not within the preset error threshold, it is judged as inconsistent.

[0066] Embodiment 2:

[0067] Embodiment 2 of the present invention discloses a safety helmet correct wearing detection system for a construction site using a safety helmet correct wearing detection method for a construction site, comprising:

[0068] De-reflection module: used for grayscale contrast and preprocessing based on RGB three channels to perform de-reflection processing on the image to be tested;

[0069] Identity matching and helmet wearing standard recognition module: obtain the face image to be recognized in the image to be tested after de-reflection processing, and match the identity of the face image to be recognized and the corresponding helmet wearing standard based on the face and posture feature recognition model of multi-task regression;

[0070] Helmet wearing detection module: The head image corresponding to the face image to be identified with successful identity matching is input into the helmet wearing detection model based on YOLOv4, and the helmet wearing detection result is output;

[0071] Helmet correct wearing detection module: Based on the helmet wearing detection results, the RGB three-channel pixel value comparison is used to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, and the helmet correct wearing detection results are output.

[0072] Embodiment 3:

[0073] Embodiment 3 of the present invention discloses an electronic device, including:

[0074] Memory for storing computer programs;

[0075] The processor is used to implement the steps of a method for detecting the correct wearing of a safety helmet at a construction site when executing a computer program.

[0076] Embodiment 4:

[0077] Embodiment 4 of the present invention discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for detecting the correct wearing of a safety helmet at a construction site are implemented.

[0078] The embodiment of the present invention discloses a method, system, device and storage medium for detecting the correct wearing of a helmet for a construction site. Through grayscale contrast and preprocessing based on RGB three channels, the problem of misjudgment caused by image reflection due to the large amount of reflective materials in the construction site is solved; through a face and posture feature recognition model based on multi-task regression, the identity of the face image to be identified and the corresponding helmet wearing standard are matched, and the problem that the existing technology cannot effectively judge the staff of the construction unit and may detect other personnel or objects that do not belong to the construction unit is solved, and the face recognition accuracy is further improved; through a helmet wearing detection model based on YOLOv4, multiple context features of maximum pooling at different scales are extracted and integrated, and the helmet wearing recognition accuracy is effectively improved; by using the RGB three-channel pixel value comparison, it is judged whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, and the problem of being unable to identify whether the helmets worn by different types of staff meet the wearing standards of their identities is solved. And through the above-mentioned technical features, the present invention conducts three gradual inspections, namely, the inspection of non-construction unit staff, the inspection of safety helmets not being worn, and the inspection of safety helmets not being worn correctly, which further improves the accuracy of safety helmet inspection.

[0079] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0080] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting the correct wearing of a safety helmet at a construction site, characterized in that: include: Step 1: Based on the grayscale contrast and preprocessing of the RGB three channels, the image to be tested is de-reflected; Step 2: Obtain a face image to be identified in the image to be tested after de-reflection processing, and match the identity of the face image to be identified and the corresponding helmet wearing standard based on a multi-task regression face and posture feature recognition model; Step 3: input the head image corresponding to the face image to be identified with successful identity matching into a helmet wearing detection model based on YOLOv4, and output the helmet wearing detection result; Step 4: Based on the helmet wearing detection result, use the RGB three-channel pixel value comparison to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, and output the helmet correct wearing detection result.

2. A method for detecting the correct wearing of a safety helmet for use at a construction site according to claim 1, characterized in that: In step 1, based on the grayscale contrast and preprocessing of the RGB three channels, the image to be tested is de-reflected, specifically: Decomposing the image to be tested into three channels of RGB to obtain grayscale images of the three channels of RGB respectively; Compare the grayscale values ​​of the RGB three-channel grayscale images under the same pixel, combine the pixel points of the channel grayscale image with the smallest grayscale value under each pixel, and obtain the reflective image to be calibrated; Compare the grayscale value of each pixel in the reflective image to be calibrated with a preset reflective grayscale threshold, set the grayscale value of pixels greater than the preset reflective grayscale threshold to 70 and calibrate them as white, and set the grayscale value of pixels less than or equal to the preset reflective grayscale threshold to 0 and calibrate them as black, to obtain a filtered reflective image; Subtract each pixel in the image to be tested from that in the filtered reflected image to obtain the image to be tested after de-reflection processing.

3. A method for detecting the correct wearing of a safety helmet for a construction site according to claim 1, characterized in that: In step 2, before the face and posture feature recognition model based on multi-task regression matches the identity of the face image to be identified and the corresponding helmet wearing standard, it also includes: a segmentation feature extraction network based on the image segmentation model and an image classification network to obtain face pixels and sub-face pixels to determine the face area.

4. A method for detecting the correct wearing of a safety helmet for a construction site according to claim 1, characterized in that: In step 2, the face and posture feature recognition model based on multi-task regression includes: a face feature recognition network, a face posture feature recognition network, a distance posture recognition network, and an angle posture recognition network; Inputting the face image to be recognized into the face feature recognition network and the posture feature recognition network respectively to obtain face feature information and posture feature information; The posture feature information is respectively input into the distance posture recognition network and the angle posture recognition network to obtain the distance posture information and the angle posture information under the posture feature information, and combined with the facial feature information to obtain the facial feature information corresponding to the distance posture information and the angle posture information under the posture feature information, so as to perform identity matching in the facial feature database.

5. A method for detecting the correct wearing of a safety helmet for a construction site according to claim 1, characterized in that: In step 3, the helmet wearing detection model based on YOLOv4 is specifically: Using the SPPNet network to perform multiple maximum pooling context feature extractions of different scales on the head image; The PANet network is used to perform parameter fusion on multiple feature maps of different scales output by the SPPNet network through convolution, up- and down-sampling, feature fusion, and stacking; The helmet wearing detection is performed based on the features obtained by fusion of the parameters.

6. A method for detecting the correct wearing of a safety helmet for use at a construction site according to claim 5, characterized in that: The pooling kernels of the maximum pooling include: 13x13, 9x9, 5x5, and 1x1.

7. A method for detecting the correct wearing of a safety helmet for use at a construction site according to claim 1, characterized in that: In step 4, based on the helmet wearing detection result, the RGB three-channel pixel value comparison is used to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, specifically: Using a feature extraction network to perform helmet cropping on the head image where the helmet wearing detection result is output as a helmet being worn; The RGB three-channel pixel value at the center of the cropped helmet image is obtained, and the difference is made with the RGB three-channel pixel value of the helmet wearing standard corresponding to the matched identity of the face image to be identified. If the difference is within the preset error threshold range, it is judged as consistent; if the difference is not within the preset error threshold range, it is judged as inconsistent.

8. A safety helmet correct wearing detection system for construction sites using the safety helmet correct wearing detection method for construction sites according to any one of claims 1 to 7, characterized in that: include: De-reflection module: used for grayscale contrast and preprocessing based on RGB three channels to perform de-reflection processing on the image to be tested; Identity matching and helmet wearing standard recognition module: obtains the face image to be recognized in the image to be tested after de-reflection processing, and matches the identity of the face image to be recognized and the corresponding helmet wearing standard based on the face and posture feature recognition model of multi-task regression; Helmet wearing detection module: inputs the head image corresponding to the face image to be identified with successful identity matching into the helmet wearing detection model based on YOLOv4, and outputs the helmet wearing detection result; Helmet correct wearing detection module: Based on the helmet wearing detection result, the RGB three-channel pixel value comparison is used to determine whether the helmet wearing of the face image to be identified is consistent with the helmet wearing standard corresponding to the matched identity, and the helmet correct wearing detection result is output.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of a method for detecting the correct wearing of a safety helmet at a construction site as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a method for detecting the correct wearing of a safety helmet for a construction site as described in any one of claims 1 to 7 are implemented.

Images