A plant behavior detection method and system based on image recognition
By setting up image acquisition devices in the factory area and using PFLD and YOLOv5 models to identify feature data, the automation and efficient, reliable monitoring of fatigue and smoking behavior in the factory area have been achieved. This solves the problems of insufficient efficiency and reliability in existing technologies and has a high cost-performance ratio.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- YALONG RIVER HYDROPOWER DEV CO LTD
- Filing Date
- 2024-05-30
- Publication Date
- 2026-07-21
AI Technical Summary
Existing factory behavior detection technologies are inefficient and unreliable, making it difficult to detect anomalies in a timely manner, and they are also costly.
An image recognition-based method is used to extract feature data by setting up image acquisition devices indoors and outdoors, using PFLD and YOLOv5 models to form feature sequences, judge fatigue work and smoking behavior, and issue alarm signals.
It improves the efficiency and reliability of behavior detection in the factory area, promptly detects potential safety hazards, saves computing resources, is cost-effective, and is suitable for various scenarios.
Smart Images
Figure CN118570858B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of behavior detection methods, and more specifically, to a factory behavior detection method and system based on image recognition. Background Technology
[0002] Factory behavior monitoring involves real-time monitoring and analysis of the behavior of employees within a factory to improve work efficiency, safety, and production quality.
[0003] Common factory behavior detection technologies involve identifying employees' identities, postures, and behaviors to determine if any abnormalities exist. The primary purpose of factory behavior detection is to improve factory safety and monitor employee conduct and efficiency. Especially when the goal is to ensure workplace safety, relying on supervisory inspections is inefficient, makes it difficult to detect problems promptly, and thus has low reliability. Introducing automated detection can solve these problems, but the technological costs must also be considered.
[0004] Therefore, it is necessary to develop a factory behavior detection method to improve the efficiency, reliability, and cost-effectiveness of behavior detection. Summary of the Invention
[0005] The purpose of this invention is to provide a factory behavior detection method and system based on image recognition, which can improve the efficiency, reliability and cost-effectiveness of behavior detection.
[0006] The embodiments of the present invention are achieved through the following technical solutions:
[0007] A factory area behavior detection method based on image recognition includes the following steps:
[0008] Multiple first image acquisition devices are installed in the indoor office area, and multiple second image acquisition devices are installed in the outdoor public area;
[0009] The first detection frame is periodically extracted continuously through the first image acquisition device, and the second detection frame is periodically extracted continuously through the second image acquisition device.
[0010] Feature data of the first detection frame and the second detection frame are extracted by a pre-trained object recognition model. The feature data includes face recognition results and corresponding eye feature data and cigarette feature data.
[0011] A first feature sequence is formed by acquiring feature data from multiple consecutive first detection frames, and a second feature sequence is formed by acquiring feature data from multiple consecutive second detection frames.
[0012] The first feature sequence is used to determine whether there is fatigue work behavior and whether there is smoking behavior in the work area. The second feature sequence is used to determine whether there is smoking behavior in public areas.
[0013] Preferably, the object recognition model includes the PFLD model and the YOLOv5 model;
[0014] The PFLD model is used to obtain the face recognition result and the corresponding eye feature data, and the YOLOv5 model is used to obtain the cigarette feature data. The YOLOv5 model is a relatively mature object detection model, and the PFLD model is a lightweight neural network model for facial landmark detection and facial pose estimation.
[0015] Preferably, the eye feature data includes the coordinates of a first point, a second point, a third point, a fourth point, a fifth point, and a sixth point of any eye;
[0016] The first point coordinates are the pixel coordinates of the inner corner of the eye, the fourth point coordinates are the pixel coordinates of the outer corner of the eye, the second point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the inner corner of the eye and the upper eyelid, the sixth point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the inner corner of the eye and the lower eyelid, the third point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the outer corner of the eye and the upper eyelid, and the fifth point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the outer corner of the eye and the lower eyelid.
[0017] The pixel coordinates (u, v) are coordinates based on the image coordinate system, where u represents the row number of the pixel and v represents the column number of the pixel.
[0018] Preferably, the method for determining whether fatigued work behavior exists is as follows:
[0019] Obtain the eye feature data sequence from the first feature sequence:
[0020]
[0021] in, Let be the coordinates of the first, second, third, fourth, fifth, and sixth points of the k-th eye feature data in the eye feature data sequence, where 1≤k≤n and n is the total number of eye feature data sequences.
[0022] The evaluation parameter π of the k-th eye feature data is obtained respectively. k :
[0023]
[0024] Where || is the modulo operator;
[0025] The evaluation parameter π k Arrange them from largest to smallest, and obtain the m largest evaluation parameters π.k Find the largest m evaluation parameters π k The average value is used to obtain the maximum reference value.
[0026] Obtain the difference reference value σ:
[0027]
[0028] If the evaluation parameter π k If the percentage of evaluation parameters that are less than the preset threshold is greater than α%, then fatigued work behavior is judged to exist.
[0029] Alternatively, if the difference reference value σ is greater than a preset difference threshold, then fatigued work behavior is determined to exist.
[0030] Otherwise, it is determined that there is no fatigue work behavior.
[0031] Preferably, the cigarette feature data includes the coordinates of the first long side and the coordinates of the second long side;
[0032] The first long side coordinates and the second long side coordinates are the pixel coordinates of two edge points of any long side of the recognition box;
[0033] The recognition box is a box that selects the cigarette after the cigarette is identified by the YOLOv5 model;
[0034] If no cigarette is detected, the cigarette feature data is set to (0, 0).
[0035] Preferably, the method for determining whether smoking occurs in the work area is as follows:
[0036] Obtain the eye feature data sequence from the first feature sequence:
[0037]
[0038] Obtain the cigarette feature data sequence from the first feature sequence:
[0039]
[0040] in, These are the coordinates of the first long side and the second long side of the k-th cigarette feature data in the cigarette feature data sequence, respectively.
[0041] Obtain the first distance parameter S from the k-th cigarette feature data respectively. k :
[0042]
[0043] The first distance parameter S kArrange the parameters in ascending order, find the q smallest first distance parameters, and calculate the average of the q smallest first distance parameters to obtain the minimum reference value.
[0044] like or If no smoking behavior is found in the work area, τ1 is the lower limit of the preset length ratio, and τ2 is the upper limit of the preset length ratio; otherwise, perform the following operations:
[0045] Obtain the second distance parameter L k :
[0046]
[0047]
[0048]
[0049] If the second distance parameter L k If the percentage of the second distance parameter that is less than the preset distance threshold is greater than β%, it is determined that there is smoking behavior in the work area; otherwise, it is determined that there is no smoking behavior in the work area.
[0050] Preferably, the method for determining whether smoking occurs in a public area is the same as the method for determining whether smoking occurs in a work area: obtaining an eye feature data sequence and a cigarette feature data sequence from the second feature sequence, with other steps being the same as the method for determining whether smoking occurs in a work area.
[0051] Preferably, an alarm signal is issued if there is evidence of fatigued work behavior, smoking in the work area, or smoking in public areas.
[0052] Preferably, if the face recognition result does not match any data in the face database, an alarm signal is issued.
[0053] To address the above problems, this invention also provides an image recognition-based factory area behavior detection system, applied to any of the above-described image recognition-based factory area behavior detection methods, comprising:
[0054] The acquisition module is used to set up multiple first image acquisition devices in indoor office areas and multiple second image acquisition devices in outdoor public areas;
[0055] The frame extraction module is used to periodically extract the first detection frame continuously through the first image acquisition device and periodically extract the second detection frame continuously through the second image acquisition device.
[0056] The recognition module is used to extract feature data of the first detection frame and the second detection frame respectively through a pre-trained object recognition model. The feature data includes face recognition results and corresponding eye feature data and cigarette feature data.
[0057] The sequence extraction module is used to obtain feature data of multiple consecutive first detection frames to form a first feature sequence, and to obtain feature data of multiple consecutive second detection frames to form a second feature sequence.
[0058] The behavior judgment module is used to determine whether there is fatigue work behavior and whether there is smoking behavior in the work area based on the first feature sequence, and to determine whether there is smoking behavior in the public area based on the second feature sequence.
[0059] The technical solutions of the embodiments of the present invention have at least the following advantages and beneficial effects:
[0060] This invention achieves automated factory area behavior detection through image recognition, improving the efficiency, timeliness, and reliability of detection.
[0061] This invention detects fatigue in work areas to prevent safety accidents caused by fatigue. It also detects smoking in work areas and public areas to prevent fire hazards, thus providing excellent supervision of factory safety.
[0062] This invention identifies key objects based on existing mature models, and then uses subsequent calculations to determine whether there are corresponding unsafe behaviors, thus saving computing resources as much as possible, improving computing speed, and further improving detection efficiency.
[0063] The detection area of this invention can be flexibly set, facilitating behavior detection at multiple different key locations.
[0064] This invention is reasonably designed, simple to build, has a high cost-performance ratio, and is easy to promote and implement. Attached Figure Description
[0065] Figure 1 This is a flowchart illustrating a factory behavior detection method based on image recognition provided in Embodiment 1 of the present invention.
[0066] Figure 2 This is a schematic diagram of the positional relationship of eye feature data provided in Embodiment 2 of the present invention; Detailed Implementation
[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0068] Example 1
[0069] This embodiment provides a factory area behavior detection method based on image recognition. (See attached document.) Figure 1 This includes the following steps:
[0070] Step S1: Set up multiple first image acquisition devices in the indoor office area and multiple second image acquisition devices in the outdoor public area;
[0071] Step S2: Periodically extract the first detection frame continuously through the first image acquisition device, and periodically extract the second detection frame continuously through the second image acquisition device;
[0072] Step S3: Extract feature data from the first detection frame and the second detection frame respectively using a pre-trained object recognition model. The feature data includes face recognition results and corresponding eye feature data and cigarette feature data.
[0073] Step S4: Obtain feature data from multiple consecutive first detection frames to form a first feature sequence, and obtain feature data from multiple consecutive second detection frames to form a second feature sequence;
[0074] Step S5: Determine whether there is fatigue work behavior and whether there is smoking behavior in the work area based on the first feature sequence, and determine whether there is smoking behavior in the public area based on the second feature sequence.
[0075] In this embodiment, the object recognition model includes the PFLD model and the YOLOv5 model;
[0076] The PFLD model is used to obtain the face recognition result and the corresponding eye feature data, and the YOLOv5 model is used to obtain the cigarette feature data.
[0077] The first image acquisition device in this embodiment, installed in the indoor office area, can be configured according to detection needs. For example, it can be installed in locations requiring close observation or operation of important equipment, or in areas with strict no-smoking or fire hazards, to prevent safety issues caused by negligence in critical or low-tolerance locations. Similarly, the second image acquisition device in the outdoor public area can be installed in areas with strict no-smoking or fire hazards. By using image acquisition and image recognition technologies to assist detection, efficient detection of smoking behavior and fatigue can be achieved, allowing for timely problem identification and prevention of safety accidents.
[0078] It's worth noting that, due to the facial recognition function, each person's eye-related data is processed individually when detecting fatigue, preventing any confusion. For periodic continuous data acquisition, images can be captured at regular intervals, with a configurable duration and frequency. For example, a 30-second image capture can be performed every 10 minutes, with one image captured every 0.2-0.5 seconds within this cycle.
[0079] Example 2
[0080] This embodiment is based on the technical solution of embodiment 1, and mainly provides further explanation of the methods for determining whether there is fatigue work behavior and the methods for determining whether smoking has occurred.
[0081] As a preferred embodiment, the eye feature data includes the coordinates of a first point, a second point, a third point, a fourth point, a fifth point, and a sixth point of any eye;
[0082] The first point coordinates are the pixel coordinates of the inner corner of the eye; the fourth point coordinates are the pixel coordinates of the outer corner of the eye; the second point coordinates are the pixel coordinates of the intersection of the vertical tangent line from the inner corner of the pupil and the upper eyelid; the sixth point coordinates are the pixel coordinates of the intersection of the vertical tangent line from the inner corner of the pupil and the lower eyelid; the third point coordinates are the pixel coordinates of the intersection of the vertical tangent line from the outer corner of the pupil and the upper eyelid; and the fifth point coordinates are the pixel coordinates of the intersection of the vertical tangent line from the outer corner of the pupil and the lower eyelid. (See also...) Figure 2 In the diagram, the points numbered 1-5 represent the first, second, third, fourth, and fifth points respectively. The first point is at the inner corner of the eye, and the fourth point is at the outer corner of the eye.
[0083] The pixel coordinates (u, v) are coordinates based on the image coordinate system, where u represents the row number of the pixel and v represents the column number of the pixel.
[0084] Furthermore, the method for determining whether fatigued work behavior exists is as follows:
[0085] Obtain the eye feature data sequence from the first feature sequence:
[0086]
[0087] in, Let be the coordinates of the first, second, third, fourth, fifth, and sixth points of the k-th eye feature data in the eye feature data sequence, where 1≤k≤n and n is the total number of eye feature data sequences.
[0088] The evaluation parameter π of the k-th eye feature data is obtained respectively. k :
[0089]
[0090] Where ‖.‖ is the modulo operator;
[0091] The evaluation parameter π k Arrange them from largest to smallest, and obtain the m largest evaluation parameters π. k Find the largest m evaluation parameters π k The average value is used to obtain the maximum reference value.
[0092] Obtain the difference reference value σ:
[0093]
[0094] If the evaluation parameter π k If the percentage of evaluation parameters that are less than the preset threshold is greater than α%, then fatigued work behavior is judged to exist.
[0095] Alternatively, if the difference reference value σ is greater than a preset difference threshold, then fatigued work behavior is determined to exist.
[0096] Otherwise, it is determined that there is no fatigue work behavior.
[0097] Evaluation parameter π k The size of the open eyes can be assessed. A key indicator of fatigue is increased blinking frequency or prolonged eye closure. When blinking frequency is excessively high, the overall evaluation parameter π is sampled. k Maximum reference value between China and Belgium Small differences will increase, so the difference reference value σ will increase, allowing for the setting of a difference threshold for evaluation; while a tendency towards excessively long periods of eye closure will result in an increase in the evaluation parameter π. k The proportion of evaluation parameters that are smaller than the preset parameter threshold increases, so it can be set as the evaluation parameter π. k The evaluation is based on the percentage of evaluation parameters that are less than the preset threshold, which is greater than α%. The difference threshold and α% can be obtained as empirical values through collected data.
[0098] On the other hand, the cigarette feature data includes the coordinates of the first long side and the coordinates of the second long side;
[0099] The first long side coordinates and the second long side coordinates are the pixel coordinates of two edge points of any long side of the recognition box;
[0100] The recognition box is a box that selects the cigarette after the cigarette is identified by the YOLOv5 model;
[0101] If no cigarette is detected, the cigarette feature data is set to (0, 0).
[0102] Furthermore, the method for determining whether smoking occurs in the work area is as follows:
[0103] Obtain the eye feature data sequence from the first feature sequence:
[0104]
[0105] Obtain the cigarette feature data sequence from the first feature sequence:
[0106]
[0107] in, These are the coordinates of the first long side and the second long side of the k-th cigarette feature data in the cigarette feature data sequence, respectively.
[0108] Obtain the first distance parameter S from the k-th cigarette feature data respectively. k :
[0109]
[0110] The first distance parameter S k Arrange the parameters in ascending order, find the q smallest first distance parameters, and calculate the average of the q smallest first distance parameters to obtain the minimum reference value.
[0111] First distance parameter S kIt does roughly determine the spatial distance between a person and a cigarette. There's a certain proportional relationship between the width of a person's eye and the length of a cigarette. However, when the difference in distance between the two and the acquisition device is too large, the pixel length in the image will differ significantly from the actual distance. Therefore, a distance is initially estimated based on their pixel-level lengths in the image. If the distance is too large, smoking can be directly excluded, for example, it might be placed on a table or discarded by someone else beforehand. When setting τ1 and τ2, the average length of the cigarette when it's intact and the average length when less is left need to be considered. It's important to note that this step primarily filters cases where the distance between the cigarette and the person is large. Therefore, τ1 can be set as much smaller as possible than the ratio of average eye width to average intact cigarette length from empirical data, for example, at least 30%. τ2 can be set as much larger as possible than the ratio of average eye width to average cigarette butt length, for example, at least 30%.
[0112] like or If no smoking behavior is found in the work area, τ1 is the lower limit of the preset length ratio, and τ2 is the upper limit of the preset length ratio; otherwise, perform the following operations:
[0113] Obtain the second distance parameter L k :
[0114]
[0115]
[0116]
[0117] If the second distance parameter L k If the percentage of the second distance parameter less than a preset distance threshold is greater than β%, then smoking in the work area is determined to exist; otherwise, smoking in the work area is determined not to exist. This means that the distance between the smoke and the eyes is used to determine whether smoking is occurring. β% can be set to a low value, or it can be set to appear in a certain number of frames. k If the distance is less than a preset threshold, smoking is detected. The second distance parameter L... k In other words, it is a parameter that measures the distance between a cigarette and a person's eye on a plane.
[0118] Furthermore, the method for determining whether smoking occurs in public areas is the same as the method for determining whether smoking occurs in work areas: eye feature data sequences and cigarette feature data sequences are obtained from the second feature sequence; the other steps are the same as the method for determining whether smoking occurs in work areas. In other words, the only difference here is that the data source is the second feature sequence; the specific methods for processing the eye feature data sequences and cigarette feature data sequences are consistent with the preceding steps.
[0119] As another preferred option, an alarm signal is issued if there is evidence of fatigued work behavior, smoking in the work area, or smoking in public areas.
[0120] Finally, if the face recognition result does not match any data in the face database, an alarm signal can be issued.
[0121] Example 3
[0122] This embodiment provides an image recognition-based factory area behavior detection system, applied to any of the above-described image recognition-based factory area behavior detection methods, including:
[0123] The acquisition module is used to set up multiple first image acquisition devices in indoor office areas and multiple second image acquisition devices in outdoor public areas;
[0124] The frame extraction module is used to periodically extract the first detection frame continuously through the first image acquisition device and periodically extract the second detection frame continuously through the second image acquisition device.
[0125] The recognition module is used to extract feature data of the first detection frame and the second detection frame respectively through a pre-trained object recognition model. The feature data includes face recognition results and corresponding eye feature data and cigarette feature data.
[0126] The sequence extraction module is used to obtain feature data of multiple consecutive first detection frames to form a first feature sequence, and to obtain feature data of multiple consecutive second detection frames to form a second feature sequence.
[0127] The behavior judgment module is used to determine whether there is fatigue work behavior and whether there is smoking behavior in the work area based on the first feature sequence, and to determine whether there is smoking behavior in the public area based on the second feature sequence.
[0128] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A factory area behavior detection method based on image recognition, characterized in that, Includes the following steps: Multiple first image acquisition devices are installed in the indoor office area, and multiple second image acquisition devices are installed in the outdoor public area; The first detection frame is periodically extracted continuously through the first image acquisition device, and the second detection frame is periodically extracted continuously through the second image acquisition device. Feature data of the first detection frame and the second detection frame are extracted by a pre-trained object recognition model. The feature data includes face recognition results and corresponding eye feature data and cigarette feature data. A first feature sequence is formed by acquiring feature data from multiple consecutive first detection frames, and a second feature sequence is formed by acquiring feature data from multiple consecutive second detection frames. Based on the first feature sequence, determine whether there is fatigue work behavior and whether there is smoking behavior in the work area; based on the second feature sequence, determine whether there is smoking behavior in the public area. The eye feature data includes the coordinates of the first, second, third, fourth, fifth, and sixth points of any eye; The first point coordinates are the pixel coordinates of the inner corner of the eye, the fourth point coordinates are the pixel coordinates of the outer corner of the eye, the second point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the inner corner of the eye and the upper eyelid, the sixth point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the inner corner of the eye and the lower eyelid, the third point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the outer corner of the eye and the upper eyelid, and the fifth point coordinates are the pixel coordinates of the intersection of the vertical tangent line of the pupil near the outer corner of the eye and the lower eyelid. The pixel coordinates (u, v) are coordinates based on the image coordinate system, where u represents the row number of the pixel and v represents the column number of the pixel. The method for determining whether fatigued work behavior exists is as follows: Obtain the eye feature data sequence from the first feature sequence: ; in, These are the first two sequences of eye feature data. The coordinates of the first, second, third, fourth, fifth, and sixth points of the eye feature data. , This represents the total number of eye feature data sequences. Obtain the first Evaluation parameters for individual eye feature data : ; in, For the modulo operator; The evaluation parameters Sort the parameters from largest to smallest and obtain the m largest evaluation parameters. Find the m largest evaluation parameters. The average value is used to obtain the maximum reference value. ; Obtain the difference reference value : ; If the evaluation parameters The proportion of evaluation parameters that are less than the preset parameter threshold is greater than This indicates that there is work fatigue. Or if the difference reference value If the difference exceeds a preset threshold, fatigue work behavior is identified. Otherwise, it is determined that there is no fatigue work behavior; The cigarette feature data includes the coordinates of the first long side and the coordinates of the second long side; The first long side coordinates and the second long side coordinates are the pixel coordinates of two edge points of any long side of the recognition box; The recognition box is a box that selects the cigarette after the cigarette is identified by the YOLOv5 model; If no cigarette is detected, the cigarette feature data is set to (0,0); The method for determining whether smoking occurs in the work area is as follows: Obtain the eye feature data sequence from the first feature sequence: ; Obtain the cigarette feature data sequence from the first feature sequence: ； in, The first of the cigarette feature data sequences is respectively the second. The coordinates of the first and second longest sides of the cigarette feature data; Obtain the distance from the first The first distance parameter of individual cigarette feature data : ； The first distance parameter Arrange the parameters in ascending order, find the q smallest first distance parameters, and calculate the average of the q smallest first distance parameters to obtain the minimum reference value. ; like or If no smoking occurs in the work area, then it is determined that no smoking behavior exists. This is the lower limit of the preset length ratio. Set the maximum length ratio; otherwise, perform the following operations: Obtain the second distance parameter : ; ; ; If the second distance parameter The proportion of the second distance parameter that is less than the preset distance threshold is greater than If the above conditions are met, it is determined that smoking has occurred in the work area; otherwise, it is determined that smoking has not occurred in the work area.
2. The factory area behavior detection method based on image recognition according to claim 1, characterized in that, The object recognition model includes the PFLD model and the YOLOv5 model; The PFLD model is used to obtain the face recognition result and the corresponding eye feature data, and the YOLOv5 model is used to obtain the cigarette feature data.
3. The factory area behavior detection method based on image recognition according to claim 1, characterized in that, The method for determining whether smoking occurs in public areas is the same as the method for determining whether smoking occurs in work areas: obtain eye feature data sequence and cigarette feature data sequence from the second feature sequence, and the other steps are the same as the method for determining whether smoking occurs in work areas.
4. The factory area behavior detection method based on image recognition according to claim 1, characterized in that, An alarm signal will be issued if there is evidence of fatigue at work, smoking in the work area, or smoking in public areas.
5. The factory area behavior detection method based on image recognition according to claim 1, characterized in that, If the face recognition result does not match any data in the face database, an alarm signal will be issued.
6. A factory area behavior detection system based on image recognition, applied to the factory area behavior detection method based on image recognition as described in any one of claims 1-5, characterized in that, include: The acquisition module is used to set up multiple first image acquisition devices in indoor office areas and multiple second image acquisition devices in outdoor public areas; The frame extraction module is used to periodically extract the first detection frame continuously through the first image acquisition device and periodically extract the second detection frame continuously through the second image acquisition device. The recognition module is used to extract feature data of the first detection frame and the second detection frame respectively through a pre-trained object recognition model. The feature data includes face recognition results and corresponding eye feature data and cigarette feature data. The sequence extraction module is used to obtain feature data of multiple consecutive first detection frames to form a first feature sequence, and to obtain feature data of multiple consecutive second detection frames to form a second feature sequence. The behavior judgment module is used to determine whether there is fatigue work behavior and whether there is smoking behavior in the work area based on the first feature sequence, and to determine whether there is smoking behavior in the public area based on the second feature sequence.