A Safety Visual Recognition Method for Fitness Equipment Based on Deep Learning

Abstract

The present invention discloses a safety visual recognition method for fitness equipment based on deep learning, which relates to the technical field of regional safety monitoring. The basic parameters of each fitness equipment are collected, and corresponding equipment areas are set for each fitness equipment according to the basic parameters of the fitness equipment. Video acquisition terminals are arranged to collect images of each equipment area through the video acquisition terminals, and the collected images are processed to obtain the images to be recognized of each equipment area. Safety recognition is carried out on each equipment area according to the obtained images to be recognized to judge whether there are potential safety hazards in the equipment area. By dividing different fitness equipment in the fitness area into different equipment areas and analyzing the risks of each equipment area according to the defined picture data sources, it is possible to carry out safety recognition on each equipment area, and when there are safety risks, corresponding early warnings can be issued in a timely manner to reduce the occurrence of danger or potential safety hazards.

Description

Technical Field

[0001] The present invention relates to the technical field of area security monitoring, and specifically, to a safety visual recognition method for fitness equipment based on deep learning. Background Art

[0002] A gym is a place where a large number of people gather. Users and employees may face potential risks when using and managing gym facilities, such as equipment failures. Therefore, it is crucial to enhance safety awareness and take corresponding preventive measures to ensure the safety of the gym. A gym safety risk warning system is a technical solution aimed at improving the safety of gym users and employees. This system aims to identify and mitigate potential safety risks to ensure the normal operation of the gym and the health and safety of users.

[0003] How to separately monitor different fitness equipment in the fitness area so that when there are people in the fitness area, potential risks can be timely warned is a problem we need to solve. For this purpose, a safety visual recognition method for fitness equipment based on deep learning is provided. Summary of the Invention

[0004] The purpose of the present invention is to provide a safety visual recognition method for fitness equipment based on deep learning.

[0005] The purpose of the present invention can be achieved through the following technical solutions: A safety visual recognition method for fitness equipment based on deep learning includes the following steps:

[0006] Step S1: Collect the basic parameters of each fitness equipment, and set corresponding equipment areas for each fitness equipment according to the basic parameters of the fitness equipment.

[0007] Step S2: Arrange video acquisition terminals, collect images of each equipment area through the video acquisition terminals, and process the collected images to obtain the images to be recognized in each equipment area.

[0008] Step S3: Perform safety recognition on each equipment area according to the obtained images to be recognized, and determine whether there are potential safety hazards in the equipment area.

[0009] Further, the process of collecting the basic parameters of each fitness equipment and setting corresponding equipment areas for each fitness equipment according to the basic parameters of the fitness equipment includes:

[0010] Construct a corresponding area floor plan according to the fitness area;

[0011] Generate corresponding equipment icons at corresponding positions in the area floor plan according to the locations of each fitness equipment in the fitness area.

[0012] Obtain the basic parameters of each fitness equipment, where the basic parameters of the fitness equipment include weight, length, width, and height;

[0013] Obtain the corresponding equipment influence coefficient according to the basic parameters of each fitness equipment;

[0014] Obtain the floor area of the corresponding fitness equipment according to the length and width of the obtained fitness equipment;

[0015] Then, according to the floor area of each fitness equipment and the equipment influence coefficient, set the corresponding equipment area for each fitness equipment, and obtain the area of the equipment area;

[0016] Map the obtained equipment area to the corresponding equipment icon in the area floor plan. According to the mapping result, obtain the relative position relationship of each equipment area in the area floor plan. The relative position relationship includes an intersection relationship, a containment relationship, and a separation relationship;

[0017] Summarize the fitness equipment corresponding to the equipment areas with a containment relationship to obtain the corresponding equipment set, and mark the equipment area with the largest area as the merged area corresponding to the equipment set.

[0018] Furthermore, arrange video acquisition terminals. The process of image acquisition of each equipment area by the video acquisition terminals includes:

[0019] Arrange a number of video acquisition terminals in the fitness area. Each video acquisition terminal adjusts the shooting range according to the location of the equipment area, so that the shooting range of each video acquisition terminal covers at least one equipment area. In the pictures of each equipment area taken, frame and mark each equipment area, which is recorded as the picture data source;

[0020] Obtain the attention degree Gz of the merged area;

[0021] According to the attention degree of each merged area, set the image sampling frequency f1 of the picture data source in the video acquisition terminal whose shooting range covers the merged area;

[0022] For the equipment areas with a separation relationship in the relative position relationship, set the image sampling frequency of the picture data source in the video acquisition terminal whose shooting range covers the corresponding equipment area to f2;

[0023] For the equipment areas with an intersection relationship in the relative position relationship, set the image sampling frequency of the picture data source in the video acquisition terminal whose shooting range covers the corresponding equipment area to f3;

[0024] Obtain the image data of each equipment area in real time through each video acquisition terminal.

[0025] Further, the process of processing the acquired images to obtain the images to be recognized in each instrument area includes:

[0026] Construct corresponding video data storage spaces according to the fitness instruments corresponding to each instrument area, and import the image data obtained from the video capture terminals where the picture data sources corresponding to the instrument areas of each fitness instrument are located into the corresponding video data storage spaces;

[0027] Construct a corresponding timeline for each instrument area, and map the image data obtained in the video data storage space into the timeline;

[0028] Perform image sampling on the image data mapped into the timeline at the image sampling frequency f3, obtain the corresponding image frames according to the sampling results, rasterize the obtained image frames, and perform grayscale processing on the rasterized image frames to obtain the corresponding grayscale images;

[0029] Intercept the obtained grayscale images according to the picture data sources, and only retain the image parts corresponding to the picture data sources, which are denoted as the images to be recognized.

[0030] Further, the process of performing safety recognition on each instrument area according to the obtained images to be recognized includes:

[0031] Input the obtained images to be recognized into the trained convolutional neural network model, perform feature extraction on the images to be recognized through the convolutional neural network model, recognize the fitness instruments in the images to be recognized according to the feature extraction results, and determine whether there are people entering the instrument area;

[0032] If there are no people entering, the image sampling frequency corresponding to the picture data source is defaulted to f3;

[0033] If there are people entering, reset the image sampling frequency corresponding to the picture data source according to the relative position relationship of the instrument area;

[0034] Perform safety recognition on each instrument area according to the changes of the fitness instruments in the recognized images to be recognized.

[0035] Further, the process of performing safety recognition on each instrument area according to the changes of the fitness instruments in the recognized images to be recognized includes:

[0036] Perform binarization processing on the obtained images to be recognized to obtain the corresponding binary images;

[0037] Perform pixel point marking on the obtained binary images, determine the pixel points belonging to the fitness instruments and summarize them, which are denoted as instrument connected areas;

[0038] Construct a corresponding circumscribed rectangular area according to the determined instrument connection area;

[0039] Set boundary endpoints for the boundaries of the determined circumscribed rectangular area;

[0040] According to the shooting time corresponding to the image to be recognized, associate the boundary endpoints at the same positions of the circumscribed rectangular areas of the instrument connection areas in adjacent images to be recognized, and compare the displacement deviations of the associated boundary endpoints;

[0041] Obtain the area safety factor Qa of the instrument area when there are people entering and when there are no people entering;

[0042] Set a corresponding risk warning factor Q for each instrument area;

[0043] Compare the obtained area risk factor with the set risk warning factor;

[0044] When Qa≥Q and there are people entering the instrument area, it means there are potential safety hazards for people in the instrument area, generate a personnel safety warning message, and send the personnel safety warning message to the management terminal;

[0045] When Qa≥Q and there are no people entering the instrument area, it means there are potential equipment safety hazards in the instrument area, generate an equipment maintenance warning message, and send the equipment maintenance warning message to the management terminal;

[0046] When Qa<Q, it means there are no safety hazards in the instrument area.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] By dividing different fitness instruments in the fitness area into different instrument areas, defining the picture data source of the image data obtained by the video acquisition terminal according to the divided instrument areas, and performing risk analysis on different situations of people entering and not entering in each instrument area respectively according to the defined picture data source, it is possible to perform safety identification on each instrument area, and when there are safety risks, issue corresponding warnings in a timely manner, reduce the occurrence of danger or potential safety hazards, and greatly improve the safety level of the fitness area. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings.

[0050] Figure 1This is the flowchart of the method of the present invention. Detailed implementation manners

[0051] As Figure 1 shown, a safety visual recognition method for fitness equipment based on deep learning includes the following steps:

[0052] Step S1: Collect the basic parameters of each fitness equipment, and set corresponding equipment areas for each fitness equipment according to the basic parameters of the fitness equipment;

[0053] Step S2: Arrange video acquisition terminals, collect images of each equipment area through the video acquisition terminals, and process the collected images to obtain the images to be recognized of each equipment area;

[0054] Step S3: Perform safety recognition on each equipment area according to the obtained images to be recognized, and determine whether there are potential safety hazards in the equipment area;

[0055] It should be further noted that, in the specific implementation process, the process of collecting the basic parameters of each fitness equipment and setting corresponding equipment areas for each fitness equipment according to the basic parameters of the fitness equipment includes:

[0056] Construct a corresponding area floor plan according to the fitness area;

[0057] According to the location of each fitness equipment in the fitness area, generate corresponding equipment icons at the corresponding positions in the area floor plan, and number each fitness equipment, denoted as i, where i = 1, 2,..., n, and n is an integer;

[0058] Obtain the basic parameters of each fitness equipment, where the basic parameters of the fitness equipment include weight, length, width, and height;

[0059] Denote the weight of the fitness equipment numbered i as D i , the length as L i , the width as S i , and the height as H i ;

[0060] Obtain the corresponding equipment influence coefficient according to the basic parameters of each fitness equipment, denoted as Qy i ;

[0061] Among them,

[0062] where a1 and a2 are weight coefficients respectively;

[0063] Obtain the floor area of the corresponding fitness equipment according to the obtained length and width of the fitness equipment, denoted as Zs i , Zs i= L i × S i ;

[0064] Then, according to the floor area of each fitness equipment and the equipment influence coefficient, a corresponding equipment area is set for each fitness equipment, and the area of the equipment area is denoted as Qs i ;

[0065] Among them, Qs i = (1 + Qy i ) × Zs i ;

[0066] The obtained equipment area is mapped to the corresponding equipment icon in the area floor plan. According to the mapping result, the relative position relationship of each equipment area in the area floor plan is obtained. The relative position relationship includes an intersection relationship, a containment relationship, and a separation relationship; among them, the intersection relationship means that part of the areas of two equipment areas overlap, the containment relationship means that one equipment area is completely within the range of another equipment area, and the separation relationship means that there is no overlapping part between two equipment areas;

[0067] The fitness equipment corresponding to the equipment areas with a containment relationship is summarized to obtain a corresponding equipment set, and the equipment area with the largest area among them is marked as the merged area corresponding to the equipment set.

[0068] It should be further noted that in the specific implementation process, the process of arranging video capture terminals and capturing images of each equipment area through the video capture terminals includes:

[0069] Arrange a number of video capture terminals in the fitness area. Each video capture terminal adjusts the shooting range according to the location of the equipment area, so that the shooting range of each video capture terminal covers at least one equipment area, so as to implement full - range monitoring of each equipment area. In the pictures of each captured equipment area, each equipment area is framed and marked, which is recorded as the picture data source;

[0070] The equipment areas included in each merged area are numbered, denoted as j, where j = 1, 2,..., m, and m is an integer;

[0071] The area of the equipment area numbered j is denoted as Qs j ;

[0072] The equipment influence coefficient of the fitness equipment corresponding to the equipment area numbered j is denoted as Qy j ;

[0073] Then the attention degree of this merged area is obtained, denoted as Gz;

[0074] Among them,

[0075] Among them, Qs max is the area of the merged region;

[0076] According to the attention degree of each merged region, set the image sampling frequency of the picture data source in the video acquisition terminal whose shooting range covers this merged region, denoted as f1;

[0077] For the instrument regions with a separated relative position relationship, set the image sampling frequency of the picture data source in the video acquisition terminal whose shooting range covers the corresponding instrument region as f2;

[0078] For the instrument regions with an intersecting relative position relationship, set the image sampling frequency of the picture data source in the video acquisition terminal whose shooting range covers the corresponding instrument region as f3;

[0079] where f1 > f3 > f2;

[0080] Real-time obtain the image data of each instrument region through each video acquisition terminal.

[0081] It should be further noted that in the specific implementation process, the process of processing the collected images to obtain the images to be recognized of each instrument region includes:

[0082] Construct corresponding video data storage spaces according to the fitness instruments corresponding to each instrument region, and import the image data obtained from the video acquisition terminal where the picture data source corresponding to the instrument region of each fitness instrument is located into the corresponding video data storage spaces;

[0083] Construct a corresponding timeline for each instrument region, and map the image data obtained in the video data storage space to the timeline;

[0084] Perform image sampling on the image data mapped to the timeline at the image sampling frequency f3, obtain the corresponding image frames according to the sampling results, perform rasterization processing on the obtained image frames, and perform gray processing on the rasterized image frames to obtain the corresponding gray images;

[0085] Intercept the obtained gray image according to the picture data source, and only retain the image part corresponding to the picture data source, denoted as the image to be recognized;

[0086] By pre-intercepting the image part of the picture data source in the gray image, the calculation amount for feature extraction in the subsequent image recognition process is greatly reduced, thereby improving the efficiency of image recognition;

[0087] It should be further noted that in the specific implementation process, the process of performing safety recognition on each instrument region according to the obtained images to be recognized includes:

[0088] Input the obtained image to be recognized into the trained convolutional neural network model, extract features from the image to be recognized through the convolutional neural network model, identify the fitness equipment in the image to be recognized according to the feature extraction results, and determine whether there is anyone entering the equipment area;

[0089] If no one enters, the image sampling frequency of the corresponding video data source defaults to f3;

[0090] If someone enters, reset the image sampling frequency of the corresponding video data source according to the relative position relationship of the equipment area; it should be further noted that in the specific implementation process, if the relative position relationship of the equipment area is a separated relationship, no modification is required. If the relative position relationship of the equipment area is an inclusion relationship, modify the image sampling frequency of the corresponding video data source to f1. If the relative position relationship of the equipment area is an intersection relationship, modify the image sampling frequency of the corresponding video data source to f2;

[0091] Perform safety identification on each equipment area according to the changes of the fitness equipment in the image to be recognized.

[0092] It should be further noted that in the specific implementation process, the process of performing safety identification on each equipment area according to the changes of the fitness equipment in the image to be recognized includes:

[0093] Perform binary processing on the obtained image to be recognized to obtain the corresponding binary image;

[0094] Perform pixel point marking on the obtained binary image, determine the pixel points belonging to the fitness equipment and summarize them, denoted as the equipment connected area;

[0095] Construct a corresponding circumscribed rectangle area according to the determined equipment connected area;

[0096] Set boundary endpoints for the boundary of the determined circumscribed rectangle area, label each set boundary endpoint as k, where k = 1, 2,..., s, and s is an integer;

[0097] According to the shooting time corresponding to the image to be recognized, associate the boundary endpoints at the same position of the circumscribed rectangle areas of the equipment connected areas in adjacent images to be recognized, and compare the displacement deviation of the associated boundary endpoints, denoted as Wp k ; it should be further noted that in the specific implementation process, the displacement deviation is the displacement change of the same boundary endpoints at the same position in adjacent images to be recognized;

[0098] Obtain the corresponding area safety factor, and denote the obtained area risk factor as Qa;

[0099] When no person enters the instrument area, then

[0100]

[0101] where f is the image sampling frequency of the picture data source corresponding to the instrument area, and Gz is the attention degree of the merged area where the instrument area is located. If the instrument area is not in any merged area, then Gz = 1;

[0102] When a person enters the instrument area, then

[0103]

[0104] where R is the number of people entering the instrument area, and α is the personnel interference coefficient;

[0105] Set a corresponding risk warning coefficient Q for each instrument area; it should be further noted that in the specific implementation process, the risk warning coefficients of each instrument area are different;

[0106] Compare the obtained area risk coefficient with the set risk warning coefficient;

[0107] When Qa≥Q and a person enters the instrument area, it means that there is a potential safety hazard for personnel in the instrument area, generate a personnel safety warning message, and send the personnel safety warning message to the management terminal;

[0108] When Qa≥Q and no person enters the instrument area, it means that there is a potential safety hazard for equipment in the instrument area, generate an equipment maintenance warning message, and send the equipment maintenance warning message to the management terminal;

[0109] When Qa<Q, it means that there is no safety hazard in the instrument area.

[0110] The above is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the equivalent embodiments with equivalent changes within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any modification or equivalent replacement made to the above embodiments based on the technical essence of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A fitness equipment safety visual recognition method based on deep learning, characterized in that: The following steps are involved: Step S1: collecting basic parameters of each fitness equipment, and setting a corresponding equipment area for each fitness equipment according to the basic parameters of the fitness equipment; Mapping the instrument area to the corresponding instrument icon in the area plan, and obtaining the relative position relationship of each instrument area in the area plan according to the mapping result, wherein the relative position relationship includes an intersection relationship, a containment relationship, and a separation relationship; The fitness equipment corresponding to the equipment areas with a containment relationship are summarized to obtain a corresponding equipment set, and the equipment area with the largest area is marked as a merged area corresponding to the equipment set; Step S2: Arrange a video acquisition terminal, use the video acquisition terminal to acquire images of each device area, and process the acquired images to obtain images to be identified of each device area. The process includes: Several video acquisition terminals are arranged in the fitness area. Each video acquisition terminal adjusts the shooting range according to the location of the equipment area, so that the shooting range of each video acquisition terminal covers at least one equipment area. In the picture of each equipment area, each equipment area is marked as a picture data source. Get the attention Gz of the merged area; According to the attention degree of each merged area, the image sampling frequency f1 of the picture data source in the video acquisition terminal whose shooting range covers the merged area is set; For the device areas whose relative position relationship is a separated relationship, the image sampling frequency of the picture data source in the video acquisition terminal whose shooting range covers the corresponding device area is set to f2; For the device areas whose relative position relationship is an intersection relationship, the image sampling frequency of the picture data source in the video acquisition terminal whose shooting range covers the corresponding device area is set to f3; Where f1>f3>f2; Acquire image data of each device area in real time through each video acquisition terminal; Step S3: Based on the obtained image to be identified, determine whether there is a person entering the equipment area, perform safety identification on each equipment area, and determine whether there is a safety hazard in the equipment area.

2. A fitness equipment safety visual recognition method based on deep learning according to claim 1, characterized in that: The process of collecting basic parameters of each fitness equipment and setting corresponding equipment areas for each fitness equipment according to the basic parameters of the fitness equipment includes: Construct a corresponding regional plan according to the fitness area; According to the location of each fitness equipment in the fitness area, a corresponding equipment icon is generated at the corresponding location in the area plan; Obtaining basic parameters of each fitness equipment, wherein the basic parameters of the fitness equipment include weight, length, width and height; According to the basic parameters of each fitness equipment, the corresponding equipment influence coefficient is obtained; According to the obtained length and width of the fitness equipment, the floor area of ​​the corresponding fitness equipment is obtained; Then, according to the floor area of ​​each fitness equipment and the equipment influence coefficient, a corresponding equipment area is set for each fitness equipment, and the area of ​​the equipment area is obtained.

3. A fitness equipment safety visual recognition method based on deep learning according to claim 2, characterized in that: The process of processing the acquired images to obtain the images to be identified in each device area includes: Construct corresponding video data storage spaces according to the fitness equipment corresponding to each equipment area, and import image data obtained by the video acquisition terminal where the picture data source corresponding to the equipment area of ​​each fitness equipment is located into the corresponding video data storage spaces; Construct a corresponding time axis for each device area, and map the image data obtained in the video data storage space to the time axis; The image data mapped to the time axis is sampled at an image sampling frequency f3, a corresponding image frame is obtained according to the sampling result, the obtained image frame is rasterized, and the rasterized image frame is gray-scaled to obtain a corresponding gray-scale image; The obtained grayscale image is intercepted according to the screen data source, and only the image portion corresponding to the screen data source is retained and recorded as the image to be recognized.

4. The method for safety visual recognition of fitness equipment based on deep learning according to claim 3, characterized in that: The process of safely identifying each device area based on the obtained image to be identified includes: Input the obtained image to be identified into a trained convolutional neural network model, perform feature extraction on the image to be identified through the convolutional neural network model, identify the fitness equipment in the image to be identified based on the feature extraction result, and determine whether there is a person entering the equipment area; If no one enters, the image sampling frequency of the corresponding screen data source defaults to f3; If a person enters, the image sampling frequency of the corresponding screen data source is reset according to the relative position relationship of the device area; According to the changes of the identified fitness equipment in the image to be identified, each equipment area is safely identified.

5. The method for safety visual recognition of fitness equipment based on deep learning according to claim 4, characterized in that: According to the changes of the fitness equipment in the image to be identified, the process of safely identifying each equipment area includes: Binarization is performed on the obtained image to be identified to obtain a corresponding binary image; Mark the pixels of the binary image, determine the pixels belonging to the fitness equipment and summarize them as the equipment connected area; Constructing a corresponding circumscribed rectangular area according to the determined device connectivity area; Setting boundary endpoints for the boundary of the determined circumscribed rectangular area; According to the shooting time corresponding to the image to be identified, the boundary endpoints at the same position of the circumscribed rectangular area of ​​the device connected area in the adjacent images to be identified are associated, and the displacement deviations of the associated boundary endpoints are compared; Obtain the regional safety factor Qa of the equipment area when there are people entering and when there are no people entering; Set a corresponding risk warning coefficient Q for each equipment area; Compare the obtained regional risk coefficient with the set risk warning coefficient; When Qa≥Q and a person enters the equipment area, it means that there is a potential safety hazard in the equipment area, and a safety warning message is generated and sent to the management terminal; When Qa≥Q and no personnel enter the equipment area, it means that there are equipment safety hazards in the equipment area, and equipment maintenance warning information is generated and sent to the management personnel terminal; When Qa<Q, it means there is no safety hazard in the equipment area.

Images