Safe wearing detection method based on YOLOv7 and edge detection algorithm

Abstract

The invention discloses a safe wearing detection method based on YOLOv7 and an edge detection algorithm, and relates to the technical field of image detection, and the method comprises the steps: obtaining a safe wearing image data set, and training a YOLOv7 detection model through the safe wearing image data set; judging whether the to-be-detected object completely enters the detection picture or not through an edge detection algorithm, and acquiring a current image frame as model input data when judging that the to-be-detected object completely enters the detection picture; inputting the model input data into the YOLOv7 detection model, and judging whether the to-be-detected object meets the safety wearing requirement or not. According to the invention, through combination of YOLOv7 and an edge detection algorithm, the accuracy and robustness of worker safe wearing detection are significantly improved, and false detection caused by the fact that a worker does not completely enter a picture is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of image detection, and particularly to a safety wear detection method based on YOLOv7 and edge detection algorithms. Background Art

[0002] With the rapid development of the new generation of information technology, emerging fields such as industrial Internet and industrial big data have continuously promoted the intelligent upgrading of the manufacturing industry. In this context, conducting research on intelligent detection and early warning of workers' safety wear based on object detection algorithms and edge detection technologies not only conforms to the development trend of artificial intelligence and Industry 4.0 but also meets the current requirements of work safety and high-quality development. The industrial working environment is complex and diverse, and the correct wearing situation of workers' wearables is directly related to production safety and workers' health. Therefore, the automatic identification and early warning of safety wear have gradually become the focus of attention of enterprises. Traditional safety wear detection methods mainly rely on manual inspections and image recognition algorithms. However, the manual method has high costs and low efficiency, while the recognition algorithms based on simple image features are vulnerable to environmental illumination, occlusion, etc. in complex scenarios, with high false detection and missed detection rates, and it is difficult to meet the requirements of the actual production environment.

[0003] With the development of computer vision and deep learning, object detection technology has made remarkable progress. In particular, the YOLO (You Only Look Once) series of models have shown good detection performance in various visual detection tasks. The YOLOv7 model provides reliable technical support for the real-time detection of safety wear through faster inference speed and higher detection accuracy. However, the YOLOv7 model may still have problems of insufficient accuracy and false detection when detecting objects at the edge of the detection screen or objects with blurred edges. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a safety wear detection method based on YOLOv7 and edge detection algorithms.

[0005] To solve the above technical problems, the technical solution of the present invention is as follows: A safety wear detection method based on YOLOv7 and edge detection algorithms, comprising: Obtaining a safety wear image dataset and training a YOLOv7 detection model through the safety wear image dataset; Judging whether the object to be detected has completely entered the detection screen through an edge detection algorithm, and obtaining the current image frame as the model input data when it is judged that the object to be detected has completely entered the detection screen; Inputting the model input data into the YOLOv7 detection model to judge whether the object to be detected meets the safety wear requirements.

[0006] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm of the present invention, wherein: the obtaining of the safety wear image dataset and the training of the YOLOv7 detection model through the safety wear image dataset include: Collect images of the detection object wearing labor protection equipment safely; Perform data augmentation and preprocessing on the collected images to construct a safety wear image dataset; Use the safety wear image dataset as sample data and input it into the YOLOv7 detection model to train the model.

[0007] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm of the present invention, wherein: the determining whether the object to be detected has completely entered the detection screen through the edge detection algorithm includes: Determine the position of the object to be detected in the detection screen through the object detection frame; Determine the position of the object detection frame and judge whether the object to be detected is at the edge of the detection screen. If not, it means that the object to be detected has completely entered the detection screen. If so, proceed to the next step; Crop the image area of the object to be detected from the detection frame area to generate a sub-image; Perform edge contour extraction on the sub-image through the edge detection algorithm to obtain an edge feature map; Perform edge contour detection on the edge feature map and judge whether the detected edge contour is closed. If so, it means that the object to be detected has completely entered the detection screen. If not, it means that the object to be detected has not completely entered the detection screen.

[0008] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm of the present invention, wherein: the determining the position of the object detection frame and judging whether the object to be detected is at the edge of the detection screen includes: Calculate the distance from the upper edge of the object detection frame to the top of the detection screen image ; Calculate the distance from the lower edge of the object detection frame to the bottom of the detection screen image ; Calculate the distance from the left edge of the object detection frame to the left end of the detection screen image ; Calculate the distance from the right edge of the object detection frame to the right end of the detection screen image ; If , it is determined that the object to be detected is at the upper edge of the detection screen. If , it is determined that the object to be detected is at the left edge of the detection screen. If , it is determined that the object to be detected is at the lower edge of the detection screen. If , it is determined that the object to be detected is at the right edge of the detection screen, where H is the height of the detection screen and W is the width of the detection screen.

[0009] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm according to the present invention, wherein: the edge contour extraction of the sub-image by the edge detection algorithm to obtain the edge feature map includes: Performing Gaussian filtering on the sub-image, and the Gaussian filtering formula is: , where is the Gaussian kernel; Using the Sobel operator to calculate the gradient of the processed sub-image to obtain the gradient magnitude G and the gradient direction , and suppressing non-edge points along the gradient direction, setting non-local maximum values to zero; Setting a high threshold and a low threshold , marking the points where the gradient magnitude G is greater than the high threshold as strong edges, marking the points where the gradient magnitude G is less than the high threshold and greater than the low threshold as weak edges, filtering out the points where the gradient magnitude G is less than the low threshold , and connecting the weak edge and strong edge regions to form an edge feature map.

[0010] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm according to the present invention, wherein: the gradient calculation of the processed sub-image to obtain the gradient magnitude G and the gradient direction includes: Calculating the gradient magnitude G through the formula , and calculating the gradient direction through the formula , where is the gradient of the brightness change rate in the horizontal direction, is the gradient of the brightness change rate in the vertical direction.

[0011] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm according to the present invention, wherein: the edge contour detection of the second target image and determining whether the detected edge contour is closed includes: Calculating the perimeter and area ratio , ; Taking the ratio and the contour closure determination threshold Compare. If , it is determined that the edge contour is closed.

[0012] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm described in the present invention, wherein: in the step of comparing the ratio with the contour closure determination threshold Compare. If , after it is determined that the edge contour is closed, it further includes: If , calculate the compactness of the edge contour. , if the difference between the compactness of the edge contour and 1 is less than the threshold, it is determined that the edge contour is closed. If the difference between the compactness of the edge contour and 1 is greater than the threshold, it is determined that the edge contour is not closed.

[0013] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm described in the present invention, wherein: inputting the model input data into the YOLOv7 detection model to determine whether the object to be detected meets the safety wear requirements includes: Judge whether all labor protection equipment is included in the current image frame through the YOLOv7 detection model. If so, it means that the object to be detected meets the safety wear requirements. If not, proceed to the next step; Obtain the next image frame in which the object to be detected completely enters the detection screen as the model input data, and input it into the YOLOv7 detection model to determine whether the object to be detected meets the safety wear requirements. If so, it means that the object to be detected meets the safety wear requirements. If not, proceed to the next step; Loop and execute the previous step until it is determined whether the object to be detected meets the safety wear requirements or it is impossible to collect the next image frame in which the object to be detected completely enters the detection screen.

[0014] As a preferred solution of the safety wear detection method based on YOLOv7 and edge detection algorithm described in the present invention, wherein: after it is impossible to collect the next image frame in which the object to be detected completely enters the detection screen, it further includes: Trigger the alarm mechanism and record all model input data.

[0015] The beneficial effects of the present invention are: (1) By combining YOLOv7 and the edge detection algorithm, the present invention significantly improves the accuracy and robustness of the worker safety wear detection, and effectively reduces the false detection caused by the worker not completely entering the screen.

[0016] (2) The introduction of the edge detection algorithm in the present invention can effectively capture the edge features of the target, thereby enhancing the recognition ability of the model in complex scenarios and effectively reducing the false detection rate and the missed detection rate.

[0017] (3) While ensuring the detection accuracy, the present invention improves the real-time performance of the system and is applicable to safety management and automatic detection in industrial environments. Brief Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0019] Figure 1 It is a schematic flowchart of the safety wear detection method based on YOLOv7 and edge detection algorithm provided by the present invention; Figure 2 It is a flowchart of the Canny edge detection algorithm provided by the present invention; Figure 3 It is an architecture diagram of the combination of YOLOv7 and edge detection algorithm provided by the present invention. Specific Embodiments

[0020] To make the content of the present invention easier to be clearly understood, the present invention will be further described in detail below according to specific embodiments in combination with the drawings.

[0021] Figure 1 It is a schematic flowchart of the safety wear detection method based on YOLOv7 and edge detection algorithm provided by the embodiments of the present application. The detection method specifically includes the following steps: Step S101: Obtain a safety wear image dataset and train a YOLOv7 detection model through the safety wear image dataset.

[0022] Specifically, images of the worker's working area are obtained in real time by installing a monitoring camera, and images of the worker wearing labor protection equipment safely are collected. Among them, the labor protection equipment includes safety helmets, labor protection clothes, and labor protection pants. Image data is continuously collected, and the images are subjected to data enhancement and preprocessing to construct a safety wear image dataset.

[0023] Among them, the data augmentation operations include rotating, horizontally flipping, and cropping the images to expand the range of image variations, aiming to improve the generalization ability of the YOLOv7 detection model and enable the model to better identify workers and labor protection equipment. In addition, to avoid affecting the recognition accuracy due to clothing color differences under different lighting conditions, lighting enhancement is also performed by adjusting the brightness, contrast, saturation, and hue of the images. The preprocessing includes further enhancing the robustness of the images through data cutting, noise addition, etc., thereby improving the reliability of the model in complex environments.

[0024] Step S102: Use an edge detection algorithm to determine whether the object to be detected has completely entered the detection screen, and obtain the current image frame as the model input data when it is determined that the object to be detected has completely entered the detection screen; Specifically, an edge detection function is introduced based on the YOLOv7 detection model to enhance the detection ability of target edges and detailed features. Edge feature extraction is an important part of image data processing, and the integrity and accuracy of the extraction of edge feature data of image data are related to the effect of image applications.

[0025] By performing edge detection preprocessing on the images, use the Canny edge detection algorithm to extract the edge information of the workers' safety wear in the images, form an edge feature map, and combine it with the input image of the YOLOv7 model to form richer input features. The steps and key formulas are as follows: a. Gaussian filtering: First perform Gaussian blur on the image I(x, y) to reduce noise interference. The formula for Gaussian filtering is: (1) where is the Gaussian kernel.

[0026] b. Gradient calculation: Use the Sobel operator to calculate the gradient magnitude G and gradient direction θ. In the Canny edge detection process, the Sobel operator is mainly used to calculate the first-order derivatives of the image in two directions (brightness changes in the horizontal and vertical directions), and the gradient information of the image is obtained through this process. The Sobel operator calculates the gradients of the image in the horizontal ( ) and vertical directions ( ) to obtain the rate and direction information of brightness changes. According to the horizontal and vertical gradients calculated by the Sobel operator, use the following formulas to obtain the magnitude (edge strength) and direction of the gradient.

[0027] Gradient magnitude (edge strength): (2) Gradient direction: (3) The Sobel operator extracts the edge structure of the image through the above method. By calculating the magnitude and direction of the gradient, the Canny algorithm can locate potential edges.

[0028] c. Non-maximum suppression: Suppress non-edge points along the gradient direction and set non-local maximum values to zero.

[0029] d. Double-threshold detection and edge connection: Set a high threshold and a low threshold , mark the points where the gradient magnitude G is greater than the high threshold as strong edges, mark the points where the gradient magnitude G is less than the high threshold and greater than the low threshold as weak edges, filter out the points where the gradient magnitude G is less than the low threshold , and connect the weak edge and strong edge regions.

[0030] Through the above steps, the Canny algorithm can effectively extract edge information, which is suitable for refining the object contour. This edge information can be combined with the feature map of YOLOv7 to help the YOLOv7 model more clearly distinguish the contours of workers and safety wear, making the model pay more attention to the key regions in the image and improving the detection accuracy of detecting objects and edges.

[0031] Based on the above content, when judging whether the object to be detected has completely entered the detection screen, first determine the position of the object to be detected in the detection screen through the object detection frame. See Figure 3 . In the figure, the rectangular frame line is the object detection frame, and the object to be detected is a worker. After the worker enters the detection screen, the object detection frame will automatically select the area where the worker is located. Then, determine the position of the object detection frame and judge whether the object to be detected is at the edge of the detection screen. The judgment method is as follows: The distance from the upper edge of the detection frame to the top of the screen image: (4) The distance from the lower edge of the detection frame to the bottom of the screen image: (5) The distance from the left edge of the detection frame to the left end of the screen image: (6) The distance from the right edge of the detection frame to the right end of the screen image: (7) Among them, H is the height of the detection screen, and W is the width of the detection screen.

[0032] According to these calculated distances, they can be compared with two preset thresholds 、 . Among them the threshold is used to judge ( , Whether it is on the screen or at the left edge, The threshold is used to judge ( , ) whether it is at the bottom or right edge of the screen.

[0033] The method for judging whether the worker detection box is close to the image edge is as follows: If , it is determined that the object to be detected is at the upper edge of the detection screen; If , it is determined that the object to be detected is at the left edge of the detection screen; If , it is determined that the object to be detected is at the lower edge of the detection screen; If , it is determined that the object to be detected is at the right edge of the detection screen.

[0034] After the above two situations can initially determine that the worker is at the edge of the screen, while performing the edge feature map judgment, the information of the actual edge features is used to improve the accuracy of the judgment.

[0035] When initially determining that the worker is at the edge of the screen through the object detection box, the image area of the worker is cropped from the object detection box area to generate a sub-image. Then, the Canny edge detection algorithm is used to perform edge contour extraction on the sub-image within the detection box to obtain the edge feature map of the worker detection box area, marked as . This image only contains the edge features of the sub-image at the position of the detection box where the worker is located, excluding other background interferences. On the edge feature map of the worker, contour detection is used to identify the edge contour of the worker, obtaining a set of closed or non-closed edge contour lines.

[0036] After obtaining the edge contour line of the worker, a closedness check is performed on the detected edge contour to determine whether the human contour is closed. If the edge contour is a closed shape and does not break at the edge of the detection box, it means that the worker has completely entered the detection box. If the contour breaks at the edge of the detection box, it means that the worker may be partially outside the screen. For the closedness of the contour, the ratio of the perimeter to the area is used to judge the degree of closure. A closed contour usually has a smaller ratio. If , it is considered that the edge contour is closed; if , it may be non-closed. Among them, is the threshold for judging the contour closure, which can be fine-tuned through data in different scenarios.

[0037] (8) Next, according to the compactness of the edge contour Further assist in determining the closure of the contour. The contour closure is checked by calculating whether the contour shape is tight and has no breaks, and the judgment formula is as follows: (9) If the compactness of the edge contour The difference from 1 is less than the threshold, that is, it is very close to 1, then it is determined that the edge contour is closed. If the compactness of the edge contour The difference from 1 is greater than the threshold, that is, it is significantly lower than 1, then it is determined that the edge contour is not closed, that is, the worker has not fully entered the detection screen.

[0038] By detecting the worker's contour through the above method, the workers with incomplete contours within the object detection frame are determined as not fully entering the screen, reducing the false detection caused by the workers not fully entering the detection screen.

[0039] After determining that the object to be detected has fully entered the detection screen, the current image frame is obtained as the model input data.

[0040] To better meet the scene requirements, in this embodiment, the target area detection and worker edge detection functions are added on the basis of YOLOv7, and the area detection can be realized by passing the detection area coordinates through the interactive interface. During the detection process, to ensure the correspondence between the worker and the labor protection clothing, two detection processes are adopted, that is, first detect the worker. After detecting a worker in the screen each time, the edge detection of the worker detection frame in the current frame of picture will be performed. When it is ensured that the worker enters the screen, this frame of picture is used as the model input data.

[0041] Step S103: Input the model input data into the YOLOv7 detection model to determine whether the object to be detected meets the safety wearing requirements.

[0042] Specifically, after the model input data is input into the YOLOv7 detection model, the YOLOv7 detection model determines whether all labor protection equipment is included in the current image frame, that is, whether the three types of labor protection equipment, namely safety helmets, labor protection clothing, and labor protection pants, exist at the same time. If so, it means that the object to be detected, that is, the worker, meets the safety wearing requirements. If not, the next step is performed; Obtain the next image frame when the object to be detected has fully entered the detection screen as the model input data, and input it into the YOLOv7 detection model to determine whether the object to be detected meets the safety wearing requirements. If so, it means that the object to be detected meets the safety wearing requirements. If not, the next step is performed; Loop and execute the previous step until it is determined whether the object to be detected meets the requirements for safe wearing or it is impossible to collect the next image frame in which the object to be detected fully enters the detection screen. If it is impossible to collect the next image frame in which the object to be detected fully enters the detection screen, it means that the worker does not meet the requirements for safe wearing. In this case, the alarm mechanism is triggered and all model input data is recorded.

[0043] It can be understood that in the above technical solution, if the three types of labor protection equipment are not detected simultaneously, the next frame will be continuously detected, which can avoid false alarms caused by missed detections due to model accuracy or scene interference.

[0044] Thus, the technical solution of this application, by combining YOLOv7 with the edge detection algorithm, significantly improves the accuracy and robustness of worker safety wearing detection, and effectively reduces false detections caused by workers not fully entering the screen.

[0045] In addition to the above embodiments, the present invention may have other implementation manners; all technical solutions formed by equivalent replacement or equivalent transformation fall within the protection scope required by the present invention.

Claims

1. A safe wear detection method based on YOLOv7 and edge detection algorithm, characterized in that: include: Obtain a safety wear image dataset, and train a YOLOv7 detection model using the safety wear image dataset; Determine whether the object to be detected has completely entered the detection screen through an edge detection algorithm, and obtain the current image frame as model input data when it is determined that the object to be detected has completely entered the detection screen; Input the model input data into the YOLOv7 detection model to determine whether the object to be detected meets the safety wearing requirements.

2. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 1, characterized in that: The obtaining of the safety wear image dataset and training the YOLOv7 detection model by using the safety wear image dataset comprises: Collect images of the test subject safely wearing labor protection equipment; Perform data enhancement and preprocessing on the collected images to build a safety wear image dataset; The safety wear image dataset is used as sample data and input into the YOLOv7 detection model to train the model.

3. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 1, characterized in that: The step of determining whether the object to be detected has completely entered the detection screen by using an edge detection algorithm includes: Determine the position of the object to be detected in the detection picture through the object detection frame; Determine the position of the object detection frame and judge whether the object to be detected is at the edge of the detection screen. If not, it means that the object to be detected has completely entered the detection screen. If yes, proceed to the next step; Crop the image area of ​​the object to be detected from the detection frame area to generate a sub-image; Perform edge contour extraction on the sub-image through an edge detection algorithm to obtain an edge feature map; Perform edge contour detection on the edge feature map and determine whether the detected edge contour is closed. If so, it means that the object to be detected has completely entered the detection screen. If not, it means that the object to be detected has not completely entered the detection screen.

4. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 3, characterized in that: Determining the position of the object detection frame and judging whether the object to be detected is at the edge of the detection screen includes: Calculate the distance between the upper edge of the object detection frame and the top of the detection screen image ; Calculate the distance between the bottom edge of the object detection frame and the bottom of the detection screen image ; Calculate the distance between the left edge of the object detection frame and the left end of the detection screen image ; Calculate the distance between the right edge of the object detection frame and the right end of the detection screen image ; like , then it is determined that the object to be detected is at the upper edge of the detection screen. If , then it is determined that the object to be detected is at the left edge of the detection screen. If , then it is determined that the object to be detected is at the lower edge of the detection screen. If , it is determined that the object to be detected is at the right edge of the detection screen, where H is the height of the detection screen and W is the width of the detection screen.

5. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 4, characterized in that: The performing edge contour extraction on the sub-image by using an edge detection algorithm to obtain an edge feature map comprises: The sub-image is subjected to Gaussian filtering, and the Gaussian filtering formula is: ,in, is the Gaussian kernel; Use the Sobel operator to calculate the gradient of the processed sub-image to obtain the gradient amplitude G and gradient direction , and suppress non-edge points along the gradient direction, setting non-local maxima to zero; Set high threshold and low threshold , the gradient amplitude G is greater than the high threshold The point is marked as a strong edge, and the gradient amplitude G is less than the high threshold and greater than the lower threshold The point is marked as a weak edge, and the gradient amplitude G is less than the low threshold The points are filtered out and the weak edge and strong edge areas are connected to form an edge feature map.

6. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 1, characterized in that: The processed sub-image is subjected to gradient calculation to obtain the gradient magnitude G and the gradient direction include: By formula Calculate the gradient amplitude G, using the formula Calculate the gradient direction ,in, is the gradient of the brightness change rate in the horizontal direction, is the gradient of the brightness change rate in the vertical direction.

7. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 1, characterized in that: The performing edge contour detection on the second target image and determining whether the detected edge contour is closed comprises: Calculate the perimeter of the edge contour and area Ratio , ; The ratio Contour closure threshold For comparison, if , then the edge contour is determined to be closed.

8. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 7, characterized in that: In the ratio Contour closure threshold For comparison, if , after determining that the edge contour is closed, it also includes: like , then calculate the compactness of the edge contour , , if the compactness of the edge contour The difference between 1 and 1 is less than the threshold, then the edge contour is considered closed. If the difference with 1 is greater than the threshold, it is determined that the edge contour is not closed.

9. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 7, characterized in that: Inputting the model input data into the YOLOv7 detection model to determine whether the object to be detected meets the safety wearing requirements includes: Use the YOLOv7 detection model to determine whether the current image frame contains all the labor protection equipment. If so, it means that the object to be detected meets the safety wearing requirements. If not, proceed to the next step; The next image frame in which the object to be detected completely enters the detection screen is obtained as the model input data, and is input into the YOLOv7 detection model to determine whether the object to be detected meets the safety wearing requirements. If so, it means that the object to be detected meets the safety wearing requirements. If not, proceed to the next step; The previous step is executed repeatedly until the object to be detected meets the safety wearing requirements or the next image frame in which the object to be detected completely enters the detection screen cannot be acquired.

10. The safe wear detection method based on YOLOv7 and edge detection algorithm according to claim 9, characterized in that: After the next image frame in which the object to be detected completely enters the detection screen cannot be acquired, the method further includes: Trigger alarm mechanisms and log all model input data.

Images