A safety helmet wearing detection system for a hydropower station

Abstract

The present invention provides a safety helmet wearing detection system for a hydropower station, comprising the following steps: S1. acquiring power station monitoring video in real time and extracting video frames containing personnel; S2. identifying personnel images in the video frames, using a sliding window method and an AdaBoot detector to identify the personnel head region image, and then using a support vector machine to identify whether the personnel is wearing a safety helmet; S3. when a personnel is not wearing a safety helmet, using at least three marker points pre-set in the video monitoring interface to perform personnel positioning based on the video frame; S4. based on the personnel positioning results, using a minimum neighborhood method to associate the personnel positioning system to obtain information on personnel not wearing safety helmets, thereby completing the detection of personnel not wearing safety helmets. The present invention fully utilizes the personnel positioning systems and video-based safety helmet wearing detection systems currently installed in power facilities such as hydropower stations and substations, overcoming the problem that current safety helmet wearing detection systems are unable to identify personnel not wearing safety helmets.

Description

Technical Field

[0001] The invention belongs to the field of intelligent detection of safety helmets, and in particular relates to a safety helmet wearing detection system for a hydropower station. Background Art

[0002] With the development of the Energy Internet and smart energy systems, a large number of hydropower stations in my country have also begun a wave of intelligent construction. As a key link in my country's energy system, the safe and stable operation of hydropower stations is of paramount importance. This is especially true in the current era of new power systems with the widespread grid connection of intermittent renewable energy sources such as wind power and photovoltaics. As a flexible and adjustable power source, the safe and stable operation of hydropower stations is becoming increasingly important. As a key tool for on-site production management, the safety helmet wearing detection system has been promoted and applied in a large number of hydropower stations in China, playing a significant role in improving the safety level of hydropower stations and reducing safety accidents.

[0003] However, the current hard hat wearing detection system has limited coverage, and is usually implemented by deploying several video devices in a specific area. Therefore, there is a problem of difficulty in obtaining on-site facial information. In addition, the current hard hat wearing detection system generally only focuses on whether there is behavior of not wearing a hard hat, but lacks identification and control of the perpetrators. Chinese Patent Publication No. CN112613449A, Publication Date April 6, 2021, the name of the invention is A hard hat wearing detection and identification method and system based on video facial images. The application discloses a hard hat wearing detection system using image recognition technology. Its shortcoming is that it does not synchronously identify the personnel information of the person who is not wearing a hard hat. In addition, for some areas of the power station involving special odors, on-site operating personnel generally wear masks, which further increases the feasibility of using a limited number of video devices configured with a hard hat wearing detection system for facial recognition.

[0004] The CN213303242U abnormal behavior alarm system combines behavior recognition and positioning systems, but it requires facial recognition of personnel information. Currently, behavior recognition and positioning are generally based on personnel information, but personnel information recognition is precisely the missing link. This is especially difficult when wearing masks, etc., using facial recognition and other technologies to identify personnel information.

[0005] At the same time, the personnel positioning system, as another important means of on-site production management, has also been promoted and applied in a large number of hydropower stations in China, and basically covers major production areas such as the main and auxiliary power plants and switch stations of hydropower stations. However, it currently focuses on monitoring the trajectories of on-site workers, serving operations management, emergency command and other businesses, and has not been applied to safety behavior detection such as wearing of safety helmets. Summary of the Invention

[0006] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a hydropower station safety helmet wearing detection system, which comprehensively utilizes the existing personnel positioning system and the safety helmet wearing detection system to solve the problem of being unable to identify people who are not wearing safety helmets in current hydropower stations.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A safety helmet wearing detection system for a hydropower station comprises the following steps:

[0009] S1. Obtain power station surveillance video in real time and extract video frames containing personnel;

[0010] S2. Identify the image of a person in the video frame, use the sliding window method and AdaBoot detector to identify the image of the person's head area, and then use a support vector machine to determine whether the person is wearing a helmet;

[0011] S3. When a person is not wearing a helmet, use at least three pre-set markers within the video surveillance interface to locate the person based on the video frame.

[0012] S4. Based on the personnel positioning results, the minimum neighborhood method is used to associate the personnel positioning system to obtain the information of personnel not wearing helmets, and complete the detection of personnel not wearing helmets.

[0013] Furthermore, step S2 includes the following steps:

[0014] S21 extracts human features from the video frame, and extracts personnel images from the video frame based on the human features;

[0015] S22. Extract the head region image from the person image using the sliding window method and the AdaBoot detector;

[0016] S23 extracts head region image features from the head region image;

[0017] S24. Use a support vector machine to analyze the image features of the head area to identify whether the person is wearing a helmet.

[0018] Furthermore, step S3 includes the following steps:

[0019] S31. Set at least three marking points within the monitoring plane of each video surveillance device;

[0020] S32. Correct the video frame without a helmet based on the markers;

[0021] S33. Based on the corrected video frame, calculate the person's position coordinates and obtain the positioning result.

[0022] Furthermore, the marking points include at least a first marking point on the same horizontal plane, with three-dimensional coordinates of (x1, y1, z), a second marking point, with three-dimensional coordinates of (x2, y2, z), a third marking point, with three-dimensional coordinates of (x3, y3, z), and a fourth marking point, with three-dimensional coordinates of (x4, y4, z). The marking box is a rectangular box on the same horizontal plane, and the correction includes horizontal tilt angle calculation, bilinear difference horizontal tilt correction, vertical tilt angle calculation, and vertical tilt correction.

[0023] Furthermore, the step S4 is based on the personnel positioning result V obtained in step S3 = (V x , V y , z), associated with the personnel positioning system result data, the three-dimensional coordinate sequence of the positioning result is {L n =(L x,n , L y,n , z), n=1,2,…,N}, N is the number of people located by the personnel positioning system, and the information of people who are not wearing helmets is determined based on the D-neighborhood. The D-neighborhood represents a circle with V as the center and D as the radius. The specific judgment rules are:

[0024]

[0025] That is, the person not wearing a helmet is identified if the person in the personnel positioning system is closest to the personnel positioning result in step S3 and the closest distance d is less than D.

[0026] Furthermore, the radius D of the D-neighborhood is not less than the positioning accuracy of the personnel positioning system.

[0027] The beneficial effects of the present invention are as follows: the present invention integrates the advantages of the helmet wearing detection system and the personnel positioning system by adding marking points / marking frames in the monitoring plane of the video equipment of the helmet wearing detection system, thereby realizing the identification and positioning of the perpetrators of unsafe behaviors such as not wearing a helmet, solving the defects of the traditional helmet wearing detection system that the behavior is detected but the perpetrator is not detected, and the personnel positioning system locates the personnel but does not identify the personnel behavior, which helps the power station to carry out safety management more effectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a safety helmet wearing detection system for a hydropower station. DETAILED DESCRIPTION

[0029] The present invention is further described in detail below with reference to the accompanying drawings and implementation examples:

[0030] like Figure 1 As shown, a safety helmet wearing detection system for a hydropower station includes the following steps:

[0031] S1. Obtain power station surveillance video in real time and extract video frames containing personnel;

[0032] S2. Identify the image of a person in the video frame, use the sliding window method and AdaBoot detector to identify the image of the person's head area, and then use a support vector machine to determine whether the person is wearing a helmet;

[0033] S3. When a person is not wearing a helmet, use at least three pre-set markers within the video surveillance interface to locate the person based on the video frame.

[0034] S4. Based on the personnel positioning results, the minimum neighborhood method is used to associate the personnel positioning system to obtain the information of personnel not wearing helmets, and complete the detection of personnel not wearing helmets.

[0035] This embodiment divides the collected hydropower station monitoring video into frames and identifies video frames containing people from each frame. Currently, there are many technologies for identifying video frames containing people. In this embodiment, a mixed Gaussian (MOG or MOG2) model is used to model the background information in the monitoring video. The goal is to separate the moving foreground target from the scene, where the moving foreground target is a person, while minimizing the impact of environmental changes such as noise and shadows. The detection results of this model will provide approximate human detection search area information for human detection, thereby improving the efficiency of subsequent human detection and reducing the false detection rate. Based on the foreground detection, it is determined whether a human body exists in the current frame image and the human body area is located. When a person is detected, the person image is extracted from each frame image, and the head area of ​​the person is determined based on the person image. The specific process for determining the human head region in each image frame is as follows: Luminous Object Vision (LUV) color features, gradient histogram (HOG) features, and human shape features are extracted from each image frame. The LUV color features, gradient histogram (HOG) features, and human shape features are fused to obtain human features. Based on the extracted human features, a person image is extracted from each frame. A sliding window method and AdaBoost are used to search for the person image and locate the human region. The human shape features are extracted by inputting the foreground image into a trained convolutional neural network. Combining a pyramid approach enables human detection at different scales. Within this framework, different parts of the human body can be detected by training the AdaBoot detector on different training data and annotated human data. Since the head region is fixed relative to the human torso, the head region can be obtained from the results of human detection and localization. The width and height of the head region are defined as 1 / 3 of the width of the human region. After the head region is determined, a head region image is first captured from each frame of the video and normalized. Next, head region image features are extracted from the head region image. These features include color, texture, and shape features of the head region, which are then fused to obtain head region image features. Finally, a support vector machine (SVM) is used to identify the extracted head region image features and determine whether the person is wearing a helmet. Therefore, step S2 specifically includes the following steps:

[0036] S21 extracts human features from the video frame, and extracts personnel images from the video frame based on the human features;

[0037] S22. Extract the head region image from the person image using the sliding window method and the AdaBoot detector;

[0038] S23 extracts head region image features from the head region image;

[0039] S24. Use a support vector machine to analyze the image features of the head area to identify whether the person is wearing a helmet.

[0040] The step S3 comprises the following steps:

[0041] S31. Set at least three marking points within the monitoring plane of each video surveillance device;

[0042] S32. Correct the video frame without a helmet based on the markers;

[0043] S33. Based on the corrected video frame, calculate the person's position coordinates and obtain the positioning result.

[0044] Furthermore, in combination with the current tilt correction in the field of image processing, this embodiment considers that the marking points include at least a first marking point on the same horizontal plane, with three-dimensional coordinates of (x1, y1, z), a second marking point, with three-dimensional coordinates of (x2, y2, z), a third marking point, with three-dimensional coordinates of (x3, y3, z), and a fourth marking point, with three-dimensional coordinates of (x4, y4, z). The marking box is a rectangular box on the same horizontal plane. The correction adopts the tilt correction method commonly used in the field of image processing, including horizontal tilt angle calculation, bilinear difference horizontal tilt correction, vertical tilt angle calculation and vertical tilt shift correction.

[0045] Furthermore, the step 4 is based on the personnel positioning result V obtained in step S3 = (V x , V y , z), associated with the personnel positioning system result data, the three-dimensional coordinate sequence of the positioning result is {L n =(L x,n , L y,n , z), n=1,2,…,N}, N is the number of people located by the personnel positioning system, and the information of people who are not wearing helmets is determined based on the D-neighborhood. The D-neighborhood represents a circle with V as the center and D as the radius. The specific judgment rules are:

[0046]

[0047] That is, the person not wearing a helmet is identified if the person in the personnel positioning system is closest to the personnel positioning result in step S3 and the closest distance d is less than D.

[0048] Furthermore, the radius D of the D-neighborhood is not less than the positioning accuracy of the personnel positioning system.

[0049] The above description is only used to illustrate a specific implementation case of the present invention and is not used to limit the scope of the present invention. All equivalent changes or modifications made by those skilled in the art without departing from the spirit and principles indicated by the present invention should still be covered by the scope of the claims of the present invention.

Claims

1. A safety helmet wearing detection system for a hydropower station, characterized in that: The following steps are involved: S1. Obtain power station surveillance video in real time and extract video frames containing personnel; S2. Identify the image of a person in the video frame, use the sliding window method and AdaBoot detector to identify the image of the person's head area, and then use a support vector machine to determine whether the person is wearing a helmet; S3. When a person is not wearing a helmet, use at least three pre-set markers within the video surveillance interface to locate the person based on the video frame. S4. Based on the personnel positioning results, the minimum neighborhood method is used to associate the personnel positioning system to obtain information about personnel not wearing helmets and complete the detection of personnel not wearing helmets; The step S3 comprises the following steps: S31. Set at least three marking points within the monitoring plane of each video surveillance device; S32. Correct the video frame without a helmet based on the markers; S33. Based on the corrected video frame, the coordinates of the personnel position are calculated to obtain the positioning result; The step S4 is based on the personnel positioning result V obtained in step S3 = (V x , V y , z), associated with the personnel positioning system result data, the three-dimensional coordinate sequence of the positioning result is {L n =(L x,n , L y,n , z), n=1,2,…,N}, N is the number of people located by the personnel positioning system, and the information of people who are not wearing helmets is determined based on the D-neighborhood. The D-neighborhood represents a circle with V as the center and D as the radius. The specific judgment rules are: That is, if the person in the personnel positioning system is closest to the personnel positioning result in step S3 and the closest distance d is less than D, the identification of the person not wearing a helmet is completed, the personnel information of the person in the identification result is obtained from the personnel positioning system, and the association between the unsafe behavior image and the personnel information is completed.

2. A hydropower station helmet wearing detection system according to claim 1, characterized in that: The step S2 comprises the following steps: S21 extracts human features from the video frame, and extracts personnel images from the video frame based on the human features; S22. Extract the head region image from the person image using the sliding window method and the AdaBoot detector; S23 extracts head region image features from the head region image; S24. Use a support vector machine to analyze the image features of the head area to identify whether the person is wearing a helmet.

3. A hydropower station helmet wearing detection system according to claim 1, characterized in that: The marking points include at least a first marking point on the same horizontal plane with three-dimensional coordinates of (x1, y1, z), a second marking point with three-dimensional coordinates of (x2, y2, z), a third marking point with three-dimensional coordinates of (x3, y3, z), and a fourth marking point with three-dimensional coordinates of (x4, y4, z). The correction includes horizontal tilt angle calculation, bilinear difference horizontal tilt correction, vertical tilt angle calculation and vertical tilt correction.

4. A hydropower station helmet wearing detection system according to claim 1, characterized in that: The radius D of the D-neighborhood is not less than the positioning accuracy of the personnel positioning system.

Images