An intelligent helmet wearing detection method and system based on multi-algorithm fusion

Abstract

The present invention relates to the field of computer vision technology, and in particular to an intelligent helmet wearing detection method and system based on multi-algorithm fusion. The method comprises acquiring image data of helmet wearing; performing preprocessing based on the acquired image data, comprising image denoising on the acquired image, enhancing the image through Laplace sharpening, and adjusting the image pixel value range through normalization; performing target detection based on the preprocessed image data, extracting image features using the target detection result, and judging the correctness of helmet wearing based on the image feature extraction, comprising performing angle calculation, symmetry analysis, and wearing gap detection based on the extracted image features; and outputting a result based on the correctness judgment. The present invention comprehensively judges the correctness of helmet wearing from three aspects: angle calculation, symmetry analysis, and wearing gap detection. The multi-dimensional judgment improves the accuracy of the judgment on the correctness of helmet wearing.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to an intelligent helmet wearing detection method and system based on multi-algorithm fusion. Background Art

[0002] The safety of construction workers is paramount in numerous industrial fields, including construction, mining, and power operations. Safety helmets, essential headgear for protecting construction workers, play a crucial role in preventing accidents. Traditionally, safety helmet wear inspections rely primarily on manual inspections, a method with numerous drawbacks. On the one hand, manual inspections are inefficient, making it difficult to effectively monitor a large number of construction workers in real time. On the other hand, manual inspections are prone to oversights and subjective inaccuracies, making it impossible to ensure that every construction worker is wearing their helmet correctly.

[0003] With the continuous development of science and technology, some sensor-based helmet detection technologies have gradually emerged. However, these technologies also have certain limitations. Sensors may be affected by environmental interference, such as strong vibrations and electromagnetic interference, which can affect the accuracy of detection. Moreover, relying solely on sensors can often only detect whether the helmet is worn. The ability to judge whether the helmet is worn correctly, such as whether the chin strap is fastened, whether the helmet is worn correctly, and the fit of the helmet on the head, is limited. In addition, in the field of image recognition technology, although there have been some attempts to apply it to safety monitoring and other aspects, its application in intelligent helmet wearing correctness detection is not mature enough. Currently, there is a lack of an effective solution that can comprehensively utilize image recognition technology to accurately determine the correctness of helmet wearing from multiple dimensions. At the same time, in the construction environment, the lighting conditions are complex and changeable, and the movements and postures of construction workers vary, which brings great challenges to the detection of correct helmet wearing. At this stage, there is a need for an intelligent helmet wearing detection method and system based on multi-algorithm fusion. Summary of the Invention

[0004] In order to solve the problems of low accuracy in traditional helmet wearing detection and the detection method being easily interfered by the external environment, the present invention provides an intelligent helmet wearing detection method and system based on multi-algorithm fusion.

[0005] In a first aspect, the present invention provides an intelligent helmet wearing detection method based on multi-algorithm fusion, which adopts the following technical solutions:

[0006] An intelligent helmet wearing detection method based on multi-algorithm fusion, including:

[0007] Acquire image data of helmet wearing;

[0008] Performing preprocessing based on the acquired image data, including image denoising and image enhancement through Laplace sharpening, and adjusting the image pixel value range through normalization;

[0009] Perform object detection based on preprocessed image data, including helmet detection using the YOLO and Faster R-CNN fusion algorithms;

[0010] Utilize the target detection results to extract image features, including edge features extracted using edge detection algorithms;

[0011] The correctness of helmet wearing is judged based on image feature extraction, including angle calculation, symmetry analysis and wearing gap detection based on the extracted image features;

[0012] The results are output based on the correctness judgment.

[0013] Furthermore, the image denoising and Laplacian sharpening of the acquired image are performed, including performing weighted averaging on the image data using Gaussian filtering, determining a neighborhood with each pixel in the image as the center, multiplying the pixel value in the neighborhood with the value of the corresponding position of the Gaussian kernel and summing them to obtain a new value of the pixel point, and finally calculating the Laplacian operator of the image and adding it to the original image, thereby enhancing the edge of the image.

[0014] Furthermore, the use of the YOLO and Faster R-CNN fusion algorithm to detect the helmet includes using the YOLO algorithm to divide the input image data into multiple grids, predicting bounding boxes and corresponding category probabilities and confidences based on the grids, calculating a confidence score based on the intersection-and-union ratio of the predicted bounding box and the true bounding box, setting a confidence threshold, and determining the area where the helmet is located based on the confidence score.

[0015] Furthermore, the use of the YOLO and Faster R-CNN fusion algorithm to detect hard hats also includes inputting the area where the hard hat is located, which is screened by the YOLO algorithm, into the region proposal network of Faster R-CNN to generate candidate regions, and performing classification and bounding box regression based on the candidate regions, wherein the classification uses a cross entropy loss function and the bounding box regression uses a Smooth L1 loss function, and the performance of classification and bounding box regression is optimized by minimizing the total loss function.

[0016] Furthermore, the use of the YOLO and Faster R-CNN fusion algorithm to detect the helmet also includes obtaining a confidence score for each detection result from the YOLO and Faster R-CNN algorithms respectively, setting a weight coefficient according to the confidence score, calculating a final confidence score using the weight coefficient, setting a confidence threshold and determining the location of the area where the helmet is located according to the final confidence score.

[0017] Furthermore, the angle calculation, symmetry analysis and wearing gap detection are performed based on the extracted image features, including determining the point coordinates on the edge contour based on the acquired edge features, calculating the second-order moment matrix of the edge contour using the point coordinates, calculating the eigenvalues ​​and eigenvectors based on the matrix, taking the eigenvector with the largest eigenvalue as the main axis direction, and performing angle calculation based on the main axis direction.

[0018] Furthermore, the angle calculation, symmetry analysis and wearing gap detection based on the extracted image features also include dividing the image for determining the position of the helmet area into left and right equal parts, calculating the SSIM values ​​of the left and right parts of the image, determining the symmetry of the left and right parts of the image based on the SSIM values, using a distance transformation algorithm to calculate the shortest distance between the helmet and the head, setting a distance threshold and comparing it with the shortest distance.

[0019] The second aspect is an intelligent helmet wearing detection system based on multi-algorithm fusion, including:

[0020] The data acquisition module is configured to: acquire image data of helmet wearing;

[0021] A preprocessing module is configured to: perform preprocessing based on the acquired image data, including image denoising and image enhancement through Laplace sharpening, and image pixel value range adjustment through normalization;

[0022] The detection module is configured to: perform object detection based on the preprocessed image data, including helmet detection using the YOLO and Faster R-CNN fusion algorithm;

[0023] The feature extraction module is configured to: extract image features using the target detection results, including extracting edge features using an edge detection algorithm;

[0024] The judgment module is configured to: judge whether the helmet is worn correctly based on image feature extraction, including angle calculation, symmetry analysis and wearing gap detection based on the extracted image features;

[0025] The output module is configured to output the result according to the correctness judgment.

[0026] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for example, a method for detecting the wearing of an intelligent helmet based on fusion of multiple algorithms.

[0027] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor to implement the intelligent helmet wearing detection method based on multi-algorithm fusion.

[0028] In summary, the present invention has the following beneficial technical effects:

[0029] 1. This invention combines the fast detection capability of the YOLO algorithm and the high-precision detection capability of the Faster R-CNN algorithm to more accurately identify helmets and determine their position and size. It also reduces false detections and missed detections based on the confidence score and the final confidence score of the fusion algorithm.

[0030] 2. The present invention comprehensively utilizes a variety of advanced image processing and analysis algorithms, including Gaussian filtering, Laplace sharpening, YOLO and Faster R-CNN fusion algorithms, edge detection algorithms, etc. The combination of these algorithms enables the system to adapt to different image quality, lighting conditions and construction environments, thereby improving the robustness and adaptability of the system.

[0031] 3. The present invention comprehensively judges the correctness of helmet wearing from three aspects: angle calculation, symmetry analysis and wearing gap detection. The second-order moment calculation of the main axis direction can accurately determine the tilt angle of the helmet; the structural similarity index SSIM can analyze the symmetry of the helmet, and the distance transformation algorithm can detect the gap between the helmet and the head. The multi-dimensional judgment improves the accuracy of the judgment of the correctness of helmet wearing. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The present invention is a schematic diagram of the overall process of an intelligent helmet wearing detection method based on multi-algorithm fusion in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described in detail below with reference to the accompanying drawings.

[0034] Example 1

[0035] Reference Figure 1 , an intelligent helmet wearing detection method based on multi-algorithm fusion in this embodiment includes:

[0036] Acquire image data of helmet wearing;

[0037] Performing preprocessing based on the acquired image data, including image denoising and image enhancement through Laplace sharpening, and adjusting the image pixel value range through normalization;

[0038] Perform object detection based on preprocessed image data, including helmet detection using the YOLO and Faster R-CNN fusion algorithms;

[0039] Utilize the target detection results to extract image features, including edge features extracted using edge detection algorithms;

[0040] The correctness of helmet wearing is judged based on image feature extraction, including angle calculation, symmetry analysis and wearing gap detection based on the extracted image features;

[0041] The results are output based on the correctness judgment.

[0042] Specifically, a smart helmet wearing detection method based on multi-algorithm fusion includes the following steps:

[0043] like Figure 1 As shown, S1, obtaining image data of helmet wearing;

[0044] High-definition cameras are installed on the top and sides of the smart helmet, ensuring that the shooting angle is at a low angle. This installation method can clearly cover the construction worker's head area and fully obtain image information of the construction worker wearing the helmet. The camera collects images of the construction worker's head in real time at a low angle. The low-angle shooting allows the image to simultaneously present the helmet, chin strap and the construction worker's head, providing rich information for subsequent image processing and analysis.

[0045] S2. performing preprocessing based on the acquired image data, including performing image denoising on the acquired image, enhancing the image through Laplace sharpening, and adjusting the image pixel value range through normalization;

[0046] For the acquired helmet wearing image data, Gaussian filtering and denoising are first performed. The two-dimensional Gaussian function formula is:

[0047] ,

[0048] in, are pixel coordinates, is the standard deviation, which controls the width of the Gaussian kernel. , the new value after Gaussian filtering:

[0049] ,

[0050] in, is the original image pixel value, Determines the neighborhood size, is the Gaussian kernel in In this way, each pixel in the image is weighted averaged to remove noise interference, and then the Laplace operator of the image is calculated. The discrete form of the Laplace operator can be expressed as:

[0051] ,

[0052] in, is the pixel in the image The grayscale value of the pixel is obtained by calculating the difference between the grayscale values ​​of the pixel and its four neighboring pixels to determine the Laplace operator value of the pixel. The calculated Laplace operator is added to the original image to obtain the image after Laplace sharpening:

[0053] ,

[0054] in, is the original image pixel value, is the weight coefficient, the weight coefficient Used to control the degree of sharpening. A larger value will make the sharpening effect more obvious, but it may cause the image to be over-sharpened, with noise and artifacts. The smaller the value, the weaker the sharpening effect. Through Laplace sharpening, the edge information of the image can be enhanced, making the outline of the helmet clearer and the features more obvious. Finally, the normalization operation is performed to change the pixel value range of the original image from Linearly map to the interval [0,1] by subtracting the minimum pixel value , the pixel value range can be shifted to a non-negative range. Then, the pixel value range is scaled to the [0, 1] range by dividing it by the difference between the maximum and minimum pixel values. This normalization process makes different images comparable, thus completing image preprocessing.

[0055] S3. Perform target detection based on the preprocessed image data, including helmet detection using the YOLO and Faster R-CNN fusion algorithm;

[0056] In order to improve the accuracy and efficiency of helmet detection, the YOLO and Faster R-CNN fusion algorithm is used to detect helmets. The input image data is divided into Grids, each grid is responsible for predicting the target object that is inside it or has a large overlap with it. For each grid, multiple bounding boxes and corresponding category probabilities and confidences are predicted. The bounding box contains information about the location and size of the target object, with the center coordinates of the bounding box , width w and height h, for each bounding box, the confidence score:

[0057] ,

[0058] in, Indicates whether there is a target object in the grid, that is, a helmet. If it exists, it is 1, and if it does not, it is 0. The intersection-of-union ratio (IoU) of the predicted bounding box and the true bounding box is calculated, that is, the area of ​​the intersection of the two bounding boxes divided by the area of ​​the union. A confidence threshold is set between 0.5 and 0.9. The area where the hard hat is located is determined based on the confidence score. If the confidence score of a grid is higher than the threshold, it is considered that there is a hard hat in the grid and it is further processed as a candidate area.

[0059] The area where the hard hat is located filtered out by the YOLO algorithm is input into the region proposal network (RPN) of Faster R-CNN. Faster R-CNN consists of the region proposal network (RPN) and Fast R-CNN. RPN is used to generate candidate areas that may contain targets. Fast R-CNN classifies and regresses these candidate areas. RPN is a fully convolutional network that slides a small window on the input feature map. At each sliding position, RPN simultaneously predicts multiple candidate areas of different scales and proportions. For the candidate areas generated by RPN, classification and bounding box regression are performed. Classification is to determine whether the candidate area contains a hard hat and what type of hard hat it is. Bounding box regression is to fine-tune the position and size of the candidate area. Bounding box regression uses the Smooth L1 loss function , the Smooth L1 loss function can be expressed as:

[0060] ,

[0061] in, are the predicted bounding box parameters (center coordinates x, y, width w and height h), are the true bounding box parameters, ,

[0062] By minimizing the total loss function , to optimize the performance of classification and bounding box regression, where Expressed as a cross entropy loss function, the confidence score of each detection result is obtained from the YOLO and Faster R-CNN algorithms respectively. Assume that the confidence score of YOLO for a certain detection result is , the confidence score of Faster R-CNN is , set the weight coefficient according to the confidence score and , calculate the final confidence score , and determine the location of the area where the helmet is located based on the final confidence score.

[0063] S4. performing image feature extraction using the target detection results, including extracting edge features using an edge detection algorithm;

[0064] After obtaining the object detection results, the detected helmet area is selected as the focus of subsequent processing. First-order partial derivative finite differences are used to calculate the image's gradient magnitude and direction. The gradient magnitude reflects the severity of pixel value changes in the image, while the gradient direction indicates the direction of the change. By calculating the gradient, we can identify areas in the image with significant pixel value changes, which are likely to be object edges. For each pixel, the gradient magnitude is the square root of the sum of the squares of the horizontal and vertical gradient components, and the gradient direction is the inverse tangent of the vertical and horizontal gradient components. After calculating the gradient magnitude and direction, non-maximum suppression is performed. This step aims to refine the edge by removing pixels that are not local maxima. In terms of the gradient direction, the gradient magnitude of the current pixel is compared with that of adjacent pixels. Only local maxima are retained, and non-maxima are suppressed, resulting in clearer and more accurate edges. High and low thresholds are set. Pixels with a gradient magnitude greater than the high threshold are considered strong edges, while pixels with a gradient magnitude between the high and low thresholds are considered weak edges. The final edge image is then formed by connecting strong edge points with their adjacent weak edge points. This preserves clear edges while avoiding the complete neglect of weaker pixels that may represent true edges. The chinstrap can be separated from the background using edge features. Since chinstraps typically have distinct edge features, edge detection makes it easy to identify their position. A shape analysis algorithm, such as the Hough transform, is then used to identify the chinstrap's shape. If the chinstrap's edge features are clear and its shape is consistent with a tightened state, the chinstrap is considered tightened.

[0065] S5. Judging the correctness of helmet wearing based on image feature extraction, including angle calculation, symmetry analysis, and wearing gap detection based on the extracted image features;

[0066] By calculating the second-order moment matrix of the edge contour of the helmet in the image, the main axis direction of the helmet is determined, and then the tilt angle of the helmet is obtained. The second-order moment matrix reflects the distribution of points on the edge contour. Its eigenvalues ​​and eigenvectors can reveal the main direction information of the image. First, the edge contour of the helmet in the image is determined, and the coordinates of the points on the edge contour are set as , calculate the second-order moment matrix:

[0067] ,

[0068] Each of these items is obtained by summing the coordinates of the points on the edge contour, and then calculating the eigenvalues ​​of the matrix M and , and the corresponding eigenvectors are and The direction determined by the eigenvector with the larger eigenvalue is considered to be the main axis direction. The tilt angle of the helmet can be calculated through the angle between the main axis direction and the horizontal direction.

[0069] Set an angle threshold, such as ±5 degrees. If the calculated helmet tilt angle is within this threshold range, it can be considered that the helmet is worn correctly; if it exceeds this range, it is considered that the helmet is not worn correctly.

[0070] A symmetry analysis is then performed, dividing the helmet image into two parts, the left and right parts. The similarity between the two parts is then measured by calculating the Structural Similarity Index (SSIM). SSIM comprehensively considers the similarity of images from three aspects: brightness, contrast, and structure, and can more accurately determine the symmetry of the helmet.

[0071] Divide the helmet image into two equal parts, labeled as image x and image y, and calculate the SSIM value:

[0072] ,

[0073] in, and are the mean of image x and y, respectively, reflecting the average brightness of the image, and are the variances of image x and y, respectively, reflecting the degree of brightness change of the image, is the covariance of images x and y, reflecting the correlation between images, 、 Is a constant used to avoid the situation where the denominator is zero. The SSIM values ​​of the left and right parts are calculated. If the SSIM value is high, it means that the left and right parts are highly similar, and it can be considered that the helmet is worn correctly. If the SSIM value is low, it is considered that the helmet is not worn correctly.

[0074] Finally, the fit between the helmet and the head is judged. The helmet and head areas are separated by segmenting the preprocessed image. Then, the distance transform algorithm is used to calculate the shortest distance between the helmet and the head to judge the fit between the two. The preprocessed image is segmented to obtain a binary image. The value of the foreground pixel (helmet or head) is set to 0, and the value of the background pixel is set to infinity.

[0075] For pixel points (i, j) and (x, y), the distance between them is:

[0076] ,

[0077] Starting from the foreground pixel, the distance value is gradually propagated to the surrounding background pixels until all pixels are assigned a value. The shortest distance between the helmet and the head is calculated, and a distance threshold is set. In this embodiment, it is 5 mm. If the calculated distance is less than this threshold, it can be considered that the helmet and the head are tightly fitted; if the distance is greater than this threshold, it is considered that there is a gap between the helmet and the head.

[0078] S6. Output the result based on the correctness judgment;

[0079] Through steps S1-S5, including image preprocessing, target detection, image feature extraction and helmet wearing correctness judgment, the correctness judgment result of the helmet wearing is obtained. The smart helmet is equipped with a small speaker, which gives a sound prompt when the helmet is not worn correctly.

[0080] Example 2

[0081] This embodiment differs from the first embodiment in that it provides an intelligent helmet wearing detection system based on multi-algorithm fusion, including:

[0082] The data acquisition module is configured to: acquire image data of helmet wearing;

[0083] A preprocessing module is configured to: perform preprocessing based on the acquired image data, including image denoising and image enhancement through Laplace sharpening, and image pixel value range adjustment through normalization;

[0084] The detection module is configured to: perform object detection based on the preprocessed image data, including helmet detection using the YOLO and Faster R-CNN fusion algorithm;

[0085] The feature extraction module is configured to: extract image features using the target detection results, including extracting edge features using an edge detection algorithm;

[0086] The judgment module is configured to: judge whether the helmet is worn correctly based on image feature extraction, including angle calculation, symmetry analysis and wearing gap detection based on the extracted image features;

[0087] The output module is configured to output the result according to the correctness judgment.

[0088] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for example, a method for detecting the wearing of an intelligent helmet based on the fusion of multiple algorithms.

[0089] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor to implement an intelligent helmet wearing detection method based on multi-algorithm fusion.

[0090] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent helmet wearing detection method based on multi-algorithm fusion, characterized in that: include: Acquire image data of helmet wearing; Performing preprocessing based on the acquired image data, including image denoising and image enhancement through Laplace sharpening, and adjusting the image pixel value range through normalization; Perform object detection based on preprocessed image data, including helmet detection using the YOLO and Faster R-CNN fusion algorithms; Utilize the target detection results to extract image features, including edge features extracted using edge detection algorithms; The correctness of helmet wearing is judged based on image feature extraction, including angle calculation, symmetry analysis and wearing gap detection based on the extracted image features; Output the results based on the correctness judgment; Denoising the acquired image and enhancing the image through Laplacian sharpening includes performing weighted averaging on the image data using Gaussian filtering, determining a neighborhood centered on each pixel in the image, multiplying the pixel values ​​in the neighborhood by the value of the corresponding position of the Gaussian kernel and summing the results to obtain a new value for the pixel, and finally calculating the Laplacian operator of the image and adding it to the original image, thereby enhancing the edges of the image; The method uses the YOLO and Faster R-CNN fusion algorithm to detect helmets, including dividing the input image data into multiple grids using the YOLO algorithm, predicting bounding boxes and corresponding class probabilities and confidence levels based on the grids, calculating a confidence score based on the intersection-over-union ratio of the predicted bounding box to the true bounding box, setting a confidence threshold, and determining the area where the helmet is located based on the confidence score; The method of using the YOLO and Faster R-CNN fusion algorithm to detect hard hats also includes inputting the hard hat area filtered by the YOLO algorithm into the Faster R-CNN region proposal network to generate candidate regions, performing classification and bounding box regression based on the candidate regions, wherein the classification uses a cross entropy loss function and the bounding box regression uses a Smooth L1 loss function, and optimizing the performance of the classification and bounding box regression by minimizing the total loss function; The method of using the YOLO and Faster R-CNN fusion algorithm to detect the helmet also includes obtaining a confidence score for each detection result from the YOLO and Faster R-CNN algorithms, setting a weight coefficient according to the confidence score, calculating a final confidence score using the weight coefficient, setting a confidence threshold, and determining the location of the helmet area based on the final confidence score; The angle calculation, symmetry analysis, and wearing gap detection are performed based on the extracted image features, including determining the coordinates of points on the edge profile based on the acquired edge features, calculating the second-order moment matrix of the edge profile using the point coordinates, calculating the eigenvalues ​​and eigenvectors based on the matrix, taking the eigenvector with the largest eigenvalue as the principal axis direction, and performing angle calculation based on the principal axis direction; The method further includes performing angle calculation, symmetry analysis, and wearing gap detection based on the extracted image features, and also includes dividing the image for determining the position of the helmet area into left and right equal parts, calculating the SSIM values ​​of the left and right parts of the image, determining the symmetry of the left and right parts of the image based on the SSIM values, calculating the shortest distance between the helmet and the head using a distance transformation algorithm, setting a distance threshold, and comparing the distance with the shortest distance.

2. An intelligent helmet wearing detection system based on multi-algorithm fusion, executing the method according to claim 1, characterized in that: include: The data acquisition module is configured to: acquire image data of helmet wearing; A preprocessing module is configured to: perform preprocessing based on the acquired image data, including image denoising and image enhancement through Laplace sharpening, and image pixel value range adjustment through normalization; The detection module is configured to: perform target detection based on the preprocessed image data, including helmet detection using the YOLO and FasterR-CNN fusion algorithm; The feature extraction module is configured to: extract image features using the target detection results, including extracting edge features using an edge detection algorithm; The judgment module is configured to: judge whether the helmet is worn correctly based on image feature extraction, including angle calculation, symmetry analysis and wearing gap detection based on the extracted image features; The output module is configured to output the result according to the correctness judgment.

3. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the intelligent helmet wearing detection method based on multi-algorithm fusion as claimed in claim 1.

4. A terminal device comprising a processor and a computer-readable storage medium, wherein the processor is configured to implement various instructions; and the computer-readable storage medium is configured to store a plurality of instructions, wherein: The instructions are suitable for being loaded by a processor and executed by the intelligent helmet wearing detection method based on multi-algorithm fusion as described in claim 1.

Images