An intelligent monitoring device and monitoring method for a work site

By extracting multiple feature values ​​and recognizing behaviors through intelligent monitoring devices at the work site, combined with wireless transmission and local storage, the problem of long-term intelligent monitoring at high-risk work sites has been solved. This enables the identification and timely reporting of violations in environments without network or power, thereby improving the level of safety management.

CN123123535APending Publication Date: 2026-11-17BEIJING DANLINGYUN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510623570.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In high-risk work sites, such as construction and blasting operations, existing technologies struggle to achieve long-term intelligent monitoring, especially in environments without internet or electricity. This makes it difficult to effectively identify and report violations in a timely manner, resulting in safety hazards being difficult to detect and address promptly.

Method used

An intelligent monitoring device was designed, comprising a video access module, an intelligent analysis module, a wireless transmission module, a storage module, an upload module, and a power supply module. It utilizes a convolutional neural network for multi-feature extraction and recognition, combines FasterNet and YOLOv8s models for image enhancement and behavior recognition, and ensures data integrity and timely reporting through 4G/5G/6G wireless transmission and local storage.

Benefits of technology

It enables efficient and intelligent monitoring in environments without internet or electricity, accurately identifies violations and reports them in a timely manner, ensures data integrity, reduces the probability of safety accidents, and improves the management and control capabilities of the work site.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123535A_ABST
    Figure CN123123535A_ABST
Patent Text Reader

Abstract

The application discloses an intelligent monitoring device and method for a work site. The device comprises a video access module for accessing one or more video monitoring devices and obtaining video signals; an intelligent analysis module for structurally identifying one or more entity objects from the video signals, generating a violation alarm event and a violation picture according to the entity objects; a wireless transmission module for transmitting the violation alarm event and the violation picture to an external monitoring platform through a wireless manner; a storage module for storing the work site video, the violation alarm event and the violation picture; an uploading module for uploading the violation alarm event and the violation picture of the storage module to the external monitoring platform when network connection is detected; a display module for displaying storage status, running status and uploading status; and a power supply module for supplying power to the entire intelligent monitoring device. The application can realize intelligent monitoring of a high-risk work site.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of safety monitoring, and more particularly to an intelligent monitoring device and method for use at work sites. Background Technology

[0002] In high-risk work sites, such as construction, blasting, and demolition operations, to strengthen information-based supervision and close management loopholes, on-site monitoring equipment is typically installed to digitally record the entire work area and the entire process of the operation. For example, using surveillance cameras, video recorders, and law enforcement recorders for information-based supervision of blasting operations helps identify potential safety hazards, reduce the probability of accidents, prevent the loss of explosives, and improve the company's end-point control capabilities.

[0003] Furthermore, many work sites lack internet and electricity, which is far from meeting the long-term monitoring needs of important locations. At the same time, work site monitoring also requires intelligent upgrades. Therefore, how to achieve long-term intelligent monitoring of various work scenarios has become a problem that needs to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent monitoring device and method for use at work sites, so as to realize intelligent monitoring of high-risk work sites.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] According to one aspect of the present invention, an intelligent monitoring device for a work site is provided, comprising:

[0007] The video access module is configured to connect to one or more video surveillance devices to acquire video signals;

[0008] The intelligent analysis module is configured to perform structured identification of one or more entity objects from video signals, and generate violation alarm events and violation images based on the one or more entity objects.

[0009] The wireless transmission module is configured to wirelessly transmit the violation alarm event and violation image to an external monitoring platform.

[0010] The storage module is configured to store on-site video, violation alarms, and violation images.

[0011] The upload module is configured to upload violation alarm events and violation images from the storage module to an external monitoring platform when network connectivity is detected.

[0012] The display module is configured to display storage status, running status, and upload status;

[0013] The power supply module is configured to power the entire intelligent monitoring device.

[0014] According to one embodiment of the present invention, the intelligent analysis module includes:

[0015] A multi-feature extraction unit is configured to extract multi-dimensional feature values ​​from video image frames;

[0016] The intelligent recognition unit is configured to identify one or more entity objects based on the multidimensional feature values;

[0017] The alarm event unit is configured to generate violation alarm events and violation images based on the one or more entity objects.

[0018] According to an embodiment of the present invention, the multi-feature extraction unit uses a convolutional neural network, which includes a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is used to extract local features of the image by sliding the convolutional kernel across the input image to progressively extract features at different locations. The pooling layer averages the extracted feature values ​​within local regions to compress the features and extract the main features. The fully connected layer maps the extracted main features to scores for different categories, flattens the feature map into a one-dimensional vector, and then performs classification by learning weights and biases.

[0019] According to one embodiment of the present invention, the classification includes: safety helmet wearing detection, smoking detection, and explosion-proof work clothes.

[0020] According to one embodiment of the present invention, the features include human-shaped features, human movement features, color features on the top of the human body, smoking features, and work clothes features.

[0021] According to one embodiment of the present invention, the entity object includes a human figure object, a human head object, a human body movement object, and a clothing object.

[0022] According to one embodiment of the present invention, when the network conditions are poor or the connection to the remote monitoring platform fails, the storage module saves alarm events, images and videos on the local storage medium; when the work is completed and the network is reconnected, the upload module automatically uploads the alarm events, images and videos on the local storage medium to the remote monitoring platform.

[0023] On the other hand, the present invention also provides a monitoring method for an intelligent monitoring device for a work site, comprising the following steps:

[0024] Acquire one or more video signals from the work site;

[0025] The video signal is structured to identify one or more entity objects, and violation alarm events and violation images are generated based on the one or more entity objects.

[0026] The violation alarm and violation images are wirelessly transmitted to an external monitoring platform.

[0027] According to one embodiment of the present invention, the step of performing structured identification of video signals to identify one or more entity objects, and generating violation alarm events and violation images based on the one or more entity objects, includes:

[0028] Extracting multidimensional feature values ​​from video image frames;

[0029] One or more entity objects are identified based on the multidimensional feature values;

[0030] Generate violation alarm events and violation images based on one or more entity objects.

[0031] According to one embodiment of the present invention, the method further includes:

[0032] Using a scene recognition method based on FasterNet, we can automatically distinguish between blasting dust scenes.

[0033] Analysis of the histogram of dust images in Lab space and elimination of histogram offsets are used to enhance the images of blasting dust.

[0034] Based on the saturation estimation of the transmission map, a single image is quickly dehazed to eliminate the effects of blasting dust and restore a clear image.

[0035] The YOLOv8s model is used to extract features from the enhanced image and identify the target objects in the image.

[0036] Based on the extracted features of the target object, and combined with a pre-trained behavior classification model, the behavior of the target object is classified and identified.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] The intelligent monitoring device of this invention features a built-in power supply module and a high-performance lithium battery pack, effectively solving the problem of inconvenient power access at blasting sites. It also boasts advantages such as portability, deployment, and immediate usability. The built-in intelligent analysis module, through intelligent analysis and alarm event linkage, effectively regulates and constrains non-standard operations by personnel at blasting sites, reducing the probability of safety accidents. Built-in storage and upload modules ensure the complete preservation of blasting site operation videos even under poor network conditions, guaranteeing the traceability of blasting operation management. It can report to the remote monitoring platform in both real-time and delayed modes, ensuring both efficiency and data security and integrity. The integrated computing chip, storage chip, wireless 5G, and wired transmission modules enable the intelligent monitoring device to feature innovative designs such as integrated computing and storage, and multi-network converged transmission. In dusty blasting scenarios, image processing technology allows for more accurate identification of monitored objects, ensuring effective identification of various violations both before and after blasting. Attached Figure Description

[0039] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0040] Figure 1 The diagram illustrates an application scenario of an intelligent monitoring device according to an exemplary embodiment of the present invention.

[0041] Figure 2 The diagram shown is a schematic diagram of an intelligent monitoring device module according to an exemplary embodiment of the present invention.

[0042] Figure 3 The diagram shown is a schematic representation of an intelligent monitoring device according to an exemplary embodiment of the present invention.

[0043] Figure 4 The diagram shown is a schematic representation of an intelligent monitoring method according to an exemplary embodiment of the present invention.

[0044] Figure 5 The diagram shown is a schematic diagram of an intelligent analysis module according to an exemplary embodiment of the present invention. Detailed Implementation

[0045] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0046] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0047] In this invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one of a, b, or c" can represent: a, b, c, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0048] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0049] like Figure 1 The diagram illustrates a working scenario for an intelligent monitoring device used at a work site. In high-risk work sites, multiple video monitoring devices 1 (1A to 1N) are typically installed. These devices can be surveillance cameras, video cameras, mobile phones, or law enforcement recorders. The video monitoring devices acquire video images from the work site and transmit them to the intelligent monitoring device 2. The intelligent monitoring device 2 then sends the intelligent analysis results of the video images to a remote monitoring platform 3.

[0050] like Figure 2 As shown, a schematic diagram of the intelligent monitoring device 2 is further provided. The intelligent monitoring device 2 includes:

[0051] The video access module 201 is configured to access one or more video surveillance devices 1 to acquire video signals;

[0052] The intelligent analysis module 202 is configured to perform structured identification of one or more entity objects from video signals, and generate violation alarm events and violation images based on the one or more entity objects.

[0053] The wireless transmission module 203 is configured to wirelessly transmit the violation alarm event and violation image to the remote monitoring platform 3.

[0054] Storage module 205 is configured to store on-site video, violation alarm events, and violation images;

[0055] Upload module 206 is configured to upload violation alarm events and violation images from storage module 205 to remote monitoring platform 3 when a network connection with remote monitoring platform 3 is detected.

[0056] Display module 204 is configured to display the storage status of the storage module, the operating status of the entire intelligent monitoring device, and the upload status of the upload module.

[0057] Power supply module 207 is configured to power the entire intelligent monitoring device.

[0058] like Figure 3 As shown, a specific embodiment of the intelligent monitoring device is given. The intelligent monitoring device is manufactured as an intelligent monitoring box, including a circular outer shell and various internal modules.

[0059] The power supply module 207 has a built-in high-performance lithium battery pack to power the entire intelligent monitoring device system. The high-performance lithium battery pack includes 3 series and 6 parallel high-performance lithium batteries.

[0060] The video access module 201 connects to one or more video surveillance devices, using a TP2826 or RTL8211 video network chip. It can connect to IPC cameras that support ONVIF, RTSP, or GB28181 protocols. The video access module includes two independent network ports: the first port is used to configure the intelligent monitoring device parameters and connect to the public network, and the second port is used to connect to monitoring devices such as cameras. It integrates ONVIF and GB28181 video access protocols to ensure that the video from the camera monitoring device is connected to the intelligent monitoring device.

[0061] The display module 204 is used to display the storage status, the operating status of the intelligent monitoring device, and the status of uploading local video. It can interact with the 128*64 LCD module using ST7567 or ST7549 driver chips to display various graphics. The operating status includes platform connection status, power information, and network operating status. The storage status includes the status of stored video recordings, memory card size, and remaining memory card storage. The upload status includes the video upload progress and whether the video upload is complete.

[0062] The wireless transmission module 203 is used to transmit locally stored video from the intelligent monitoring device to the remote monitoring platform 3. This can be achieved using MINIPCI interface EC20 and 9300C model modules. The wireless transmission module integrates 4G, 5G, or 6G wireless transmission modules for transmitting video and data to the remote monitoring platform.

[0063] Storage module 205 is used to store on-site video of the operation, using AU6258 and AU6435R storage adapter chips.

[0064] The upload module 206 is used to upload the stored video recordings to the remote monitoring platform 3 for video storage when the intelligent monitoring device is connected to the public network. It uses an STM32103C8T6 and a TPS54334 switching chip to start uploading data when connected to the public network.

[0065] The intelligent analysis module 202 integrates multiple intelligent recognition algorithms for on-site operations, used for encoding and intelligent analysis of acquired videos. This module includes either the MegBoard-M1L core intelligent algorithm analysis board or the AI-58SC-RK3588-V1.0 core board, used for structured recognition of people, vehicles, and objects, and for generating events based on the recognition results, especially potential accident hazards. The generated alarm events and violation images can be uploaded to a remote monitoring platform via a wireless transmission module, or stored in a storage module to form valid records. The intelligent analysis module 202 integrates a 16T computing chip, with options including the RK3588J, BM1684, or AICore 630A.

[0066] The storage module is connected to the upload module. When the work site cannot be connected to the network, the video and data are stored in the storage module. When the network is connected, the video and data in the storage module are transmitted to the remote monitoring platform through the upload module to ensure the integrity of the recording.

[0067] The entire interface of the intelligent monitoring device is waterproof and sealed, with a protection level of IP66.

[0068] The video access module connects the video signal from the work site to the intelligent analysis module. The intelligent analysis module analyzes the acquired video using intelligent algorithms, generating alarm events, images, and videos. It also activates two parallel channels. When the 4G, 5G, or 6G network signal is good at the work site, the wireless transmission module acts as the first channel to upload alarm events, images, and videos to the remote monitoring platform. When the network condition is poor, the storage module acts as the second channel to save alarm events, images, and videos to local storage media. After the blasting operation is completed, the blasting operation smart box is taken to a location with network access, connected to the network, and the upload module automatically uploads the alarm events, images, and videos from the local storage media to the blasting operation monitoring platform, forming an effective blasting operation control recording.

[0069] like Figure 4 As shown, a monitoring method based on an intelligent monitoring device is presented, including the following steps:

[0070] Step S401: Acquire one or more video signals from the work site;

[0071] Step S402: Perform structured identification on the video signal to identify one or more entity objects, and generate violation alarm events and violation images based on the one or more entity objects; the entity objects include human-shaped objects, human-top objects, human-movement objects, and clothing objects.

[0072] Step S403: Send the violation alarm event and violation image to an external monitoring platform wirelessly.

[0073] like Figure 5 As shown, the intelligent analysis module 202 includes:

[0074] The multi-feature extraction unit 2021 is configured to extract multi-dimensional feature values ​​from video image frames;

[0075] The intelligent recognition unit 2022 is configured to identify one or more entity objects based on the multidimensional feature values;

[0076] Alarm event unit 2023 is configured to generate violation alarm events and violation images based on the one or more entity objects.

[0077] The multi-feature extraction unit 2021 uses a convolutional neural network (CNN). The CNN includes a convolutional layer 501, a pooling layer 502, and a fully connected layer 503.

[0078] The detailed implementation steps for multi-feature value extraction are as follows:

[0079] Step S1: The convolutional layer is used to extract local features of the image. The operation involves sliding the convolution kernel across the input image to progressively extract features at different locations. The algorithm calculation formula is as follows:

[0080]

[0081] X represents the image two-dimensional matrix, W represents the convolution kernel two-dimensional matrix, S represents the calculated feature matrix, and i,j,m,n represent the coordinates of the two-dimensional matrix data.

[0082] For example, in detecting whether a safety helmet is being worn, human features, top color features (hair color features, red, yellow, and white color features of the safety helmet), and human movement feature values ​​are extracted from the acquired video images.

[0083] Step S2: The pooling layer calculates the average value of the local region of the feature value extracted in step S1 to compress the feature and extract the main features.

[0084]

[0085] M(i,j) represents the (i,j)th element of the eigenvalues; ∑ represents the summation operation; f×f is the size of the pooling window; u and v are indices that vary in the range [0,f-1] and are used to traverse each element within the pooling window; s represents the step size, which defines the distance the pooling window moves over the input eigenvalues.

[0086] A(i×s+u,j×s+v) is the input feature map, representing the elements of the local window corresponding to the output feature map M(i,j).

[0087] Step S3: The fully connected layer maps the extracted features to scores for different categories, flattens the feature map into a one-dimensional vector, and then performs classification by learning weights and biases. The fully connected layer classifies the compressed feature values ​​from step S2. The classifications include: helmet wearing detection, smoking detection, and explosion-proof work clothes.

[0088] For example, in a helmet-wearing detection algorithm, several features are identified: human shape, human movement, and the color of the person's head (red, yellow, or white). Based on these features, the judgment is: helmet worn; otherwise, if the person's head is black, the judgment is: helmet not worn. The human shape feature, representing human movement, serves as an auxiliary detection method, while the head color, based on color recognition, is the core detection method. This effectively avoids misjudgments caused by algorithmic misidentification. The calculation formula is as follows.

[0089] W = P(i) * M(j) * C(k)

[0090] W represents the weight after identification. For example, if the weight for not wearing a helmet is greater than 80, it can be determined that no helmet is being worn.

[0091] i, j, k represent calculation factors, with values ​​ranging from 0 to 100%.

[0092] P(i) represents the humanoid feature value, ranging from 0 to 100. The larger the value, the more obvious the humanoid feature.

[0093] M(j) represents the judgment value of human movement characteristics, ranging from 0 to 100. The larger the value, the more obvious the distinction of movement.

[0094] C(k) represents the color feature value of the top of the human figure, ranging from 0 to 100. The larger the value, the more obvious the color distinction.

[0095] Example 1: Intelligent recognition of whether a safety helmet is being worn

[0096] Traditional feature extraction is often single-dimensional, which can easily lead to misjudgments. At blasting sites, the primary concern is ensuring blasting personnel wear safety helmets and explosion-proof protective clothing, and that smoking is not prohibited. For example, wearing a safety helmet can be used to extract multiple features from multiple dimensions, including: human shape features, color features of the person's top (hair color, red, yellow, or white helmet color), and whether the person is moving or stationary.

[0097] After extracting multidimensional feature values, a multi-step filtering process can be performed: first, filtering construction workers; second, detecting movement of the workers; and third, identifying if the top of the worker is black. Combining these three steps with multidimensional weighting, it can be determined whether a safety helmet is being worn. These three steps can be simplified to form a safety helmet wearing detection model. The model obtains the following features: worker feature P (Person), worker movement feature M (Move), top color C (Color), and safety helmet wearing weight W (Weight). The calculation formula for the safety helmet wearing detection model is as follows:

[0098] The weight of wearing a safety helmet is W = human figure feature P(i) * human figure movement feature M(i) * color of the top of the person C(i).

[0099] Where i represents the calculation factor, and P(i), M(j), and C(k) represent the calculation functions; the larger the weight W for wearing a safety helmet, the more likely it is that a safety helmet is being worn, and the smaller the weight W is, the more likely it is that a safety helmet is not being worn.

[0100] Traditional feature extraction only uses the color C(i) of the top of the person to determine whether they are wearing a helmet, which has a low recognition accuracy. Multi-feature extraction can improve the recognition accuracy of the algorithm.

[0101] As an extended application, the monitoring of whether a safety helmet is worn, as described in Example 1, can detect violations such as not wearing explosion-proof work clothes and smoking at blasting operation sites. Both can be achieved using multi-feature extraction methods, which will not be elaborated in detail here.

[0102] The intelligent recognition unit 2022 performs identification and analysis on the categories of safety helmet wearing detection, smoking detection, and explosion-proof work clothes detection classified by the multi-feature value extraction unit, and generates a work order for violation alarm events at the blasting operation site, which is sent to the alarm event unit. For example, if the intelligent recognition unit identifies and analyzes that personnel at the blasting operation site are not wearing safety helmets, it will generate a list of violation alarms for not wearing safety helmets.

[0103] The alarm event unit 2023 will generate two alarm events from the received work orders for violations at the blasting operation site. The alarm event content mainly includes the type of violation by personnel at the blasting site (not wearing a safety helmet, not wearing a blast suit, smoking in violation of regulations), violation pictures, and violation videos.

[0104] The first alarm event can be immediately pushed to the administrator or remote monitoring device via a 4G, 5G, or 6G wireless network module, facilitating timely viewing and handling, and improving the safety of the blasting operation site.

[0105] The second channel alarm event has two parts. The first part can be sent to a video monitoring terminal, such as a PTZ camera. The video monitoring terminal receives the alarm event and triggers the voice alarm function, such as the voice alarm message: Please wear your safety helmet in time, please wear your protective clothing in time, smoking is prohibited in the blasting operation area, etc., which effectively reminds the on-site blasting operation personnel to correct their violations in time. The second part is saved on the local storage medium. After the blasting operation is completed, the intelligent monitoring device is taken to a place with network access, connected to the network, and the alarm event, pictures and videos on the local storage medium are automatically uploaded to the remote monitoring platform through the upload module, forming an effective operation control recording.

[0106] Example 2: On-site application of intelligent monitoring devices

[0107] In high-risk work sites, taking blasting operations as an example, the video feed from video monitoring equipment 1, such as a surveillance sphere, is bound to the intelligent monitoring device 2. Simultaneously, the intelligent recognition algorithms of the intelligent monitoring device 2 are configured and enabled, such as helmet-wearing detection, unsafe behavior analysis, and substitute worker alarms. When work begins at the blasting site, the surveillance sphere and intelligent monitoring device 2 are placed on a tripod, the sphere is aligned with the blasting site, and the switch for both is pressed, activating the entire system.

[0108] The monitoring system, including the PTZ camera, captures and encodes video through a video access module. The captured image information is then fed to the intelligent analysis module, which integrates the MegBoard-M1L high-performance computing chip. Using a convolutional neural network (CNN), the module analyzes the images to extract characteristics of violations by personnel at the blasting operation site. A classification model is used to determine if personnel are not wearing safety helmets, protective clothing, or smoking. Upon detecting a violation, an alarm is immediately triggered, and the alarm record is stored in a server database system, including the duration, address, image, and video of the violation. Simultaneously, real-time monitoring and alerts transmit violation records and videos to a remote monitoring station for blasting operations via 4G, 5G, or 6G modules, creating an effective blasting operation control recording. The intelligent monitoring device's power supply module incorporates a high-capacity lithium battery, allowing it to operate for up to 10 hours without an external power supply. Furthermore, the entire intelligent monitoring device for blasting operations features a sealed design with an IP66 protection rating, enabling operation in harsh rainy environments.

[0109] The intelligent analysis module integrates a 16T high-performance computing chip, performs data analysis through deep learning image algorithms, generates alarm events, images, and videos, and opens two parallel channels. The wireless transmission module uploads the intelligently analyzed alarm events, images, and videos to the video monitoring platform, which can monitor the site in real time. When the network is poor or the connection to the video monitoring platform fails, the storage module saves the alarm events, images, and videos to the local storage medium. After the blasting operation is completed, the blasting operation smart box is taken to a place with network access, connected to the network, and the upload module automatically uploads the alarm events, images, and videos from the local storage medium to the blasting operation monitoring platform, forming an effective blasting operation control recording.

[0110] In summary, the power supply module of the intelligent monitoring device of the present invention can meet the requirements of long-term operation of the deployment monitoring; the intelligent monitoring device adopts a waterproof and sealed interface design, which effectively ensures product stability and meets the requirements of safe and stable operation in various harsh rainy environments; the intelligent analysis module of the intelligent monitoring device increases the computing power of the deployment monitoring product, integrates intelligent analysis algorithms, and intelligently upgrades some old deployment monitoring equipment. It is simple to operate and widely used in mobile monitoring of various complex locations at blasting operation sites.

[0111] Example 3: Automatic Behavior Recognition Based on Images in Blasting Dust Scenarios by Intelligent Analysis Module

[0112] This paper uses a scene recognition method based on FasterNet to automatically distinguish between scenes of blasting dust. The FasterNet model structure consists of four stages. Each stage contains a FasterNet Block and is connected to an embedding layer (Conv 4×4, stride 4) and a merging layer (Conv 2×2, stride 2) for spatial downsampling and channel expansion. After the four stages are completed, the features from the last three layers are used for classification.

[0113] In the data preprocessing stage, an image enhancement method for blasting dust based on the Lab color space is adopted. First, the histogram of the dust image in the Lab space is analyzed to eliminate histogram offsets, thereby restoring the original color information of the image and stretching the brightness. Then, a fast dehazing method for single images based on saturation estimation of the transmission map is introduced to enhance the image contrast and saturation, eliminate the effects of blasting dust, and restore a clear image.

[0114] The YOLOv8s model is used to extract features from the enhanced image and identify target objects in the image. A heatmap is then created using Layer-CAM (layer class activation mapping) to visually observe the model's focus on key feature regions of the target object, thus effectively identifying the object's presence and location.

[0115] Based on the extracted features of the target object, and combined with a pre-trained behavior classification model, the behavior of the target object is classified and identified. Behavior classification can be achieved using convolutional neural networks (CNNs) or other suitable classification algorithms from deep learning.

[0116] For blasting scenarios, which are often accompanied by dust pollution, this algorithm combines blasting dust scene recognition and image enhancement techniques to adapt to image behavior recognition tasks in complex environments such as sandstorms. Through data preprocessing and enhancement, it effectively addresses the impact of sandstorms on image quality during blasting operations, improving the robustness and stability of behavior recognition. Utilizing a lightweight YOLOv8s model and optimized feature extraction and analysis methods, the algorithm maintains high accuracy while exhibiting low computational complexity and latency.

[0117] Furthermore, an exemplary embodiment of the present invention may also provide a computer-readable storage medium storing a computer program. This computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform an intelligent monitoring method according to an exemplary embodiment of the present invention. The computer-readable recording medium is any data storage device capable of storing data read by a computer system. Examples of computer-readable recording media include: read-only memory, random access memory, read-only optical disc, magnetic tape, floppy disk, optical data storage device, and carrier waves (such as data transmission via the Internet through wired or wireless transmission paths).

[0118] Furthermore, an exemplary embodiment of the present invention may also provide a computing device. The computing device includes a processor and a memory. The memory stores a computer program. The computer program is executed by the processor, causing the processor to perform the intelligent monitoring method according to an exemplary embodiment of the present invention.

[0119] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and other materials. In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several of the functions listed in the specification. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0120] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. An intelligent monitoring device for a job site, characterized by, include: The video access module is configured to connect to one or more video surveillance devices to acquire video signals; The intelligent analysis module is configured to perform structured identification of one or more entity objects from video signals, and generate violation alarm events and violation images based on the one or more entity objects. The wireless transmission module is configured to wirelessly transmit the violation alarm event and violation image to an external monitoring platform. The storage module is configured to store on-site video, violation alarms, and violation images. The upload module is configured to upload violation alarm events and violation images from the storage module to an external monitoring platform when network connectivity is detected. The display module is configured to display storage status, running status, and upload status; The power supply module is configured to power the entire intelligent monitoring device.

2. The intelligent monitoring device for work sites according to claim 1, characterized in that, The intelligent analysis module includes: A multi-feature extraction unit is configured to extract multi-dimensional feature values ​​from video image frames; The intelligent recognition unit is configured to identify one or more entity objects based on the multidimensional feature values; The alarm event unit is configured to generate violation alarm events and violation images based on the one or more entity objects.

3. The intelligent monitoring device for work sites according to claim 2, characterized in that, The multi-feature extraction unit uses a convolutional neural network, which includes convolutional layers, pooling layers, and fully connected layers. Specifically: the convolutional layers extract local features of the image by sliding the convolutional kernel across the input image to progressively extract features from different locations; the pooling layers average the extracted feature values ​​within local regions to compress the features and extract the main features; and the fully connected layers map the extracted main features to scores for different categories, flattening the feature map into a one-dimensional vector, and then classifying the features by learning weights and biases.

4. The intelligent monitoring device for work sites according to claim 3, characterized in that, The categories include: safety helmet wearing detection, smoking detection, and explosion-proof work clothes.

5. The intelligent monitoring device for work sites according to claim 3, characterized in that, The features include human-shaped features, human movement features, color features on the top of the human head, smoking features, and work clothes features.

6. The intelligent monitoring device for work sites according to claim 2, characterized in that, The entity objects include humanoid objects, humanoid top objects, humanoid movement objects, and clothing objects.

7. The intelligent monitoring device for work sites according to claim 1, characterized in that, When network conditions are poor or the connection to the remote monitoring platform fails, the storage module saves alarm events, images, and videos to the local storage medium. When the operation is completed and the network is reconnected, the upload module automatically uploads the alarm events, images, and videos from the local storage medium to the remote monitoring platform.

8. A monitoring method based on the intelligent monitoring device for a work site according to claims 1 to 7, characterized in that, Includes the following steps: Acquire one or more video signals from the work site; The video signal is structured to identify one or more entity objects, and violation alarm events and violation images are generated based on the one or more entity objects. The violation alarm and violation images are wirelessly transmitted to an external monitoring platform.

9. The monitoring method according to claim 8, characterized in that, The process of structurally identifying one or more entity objects from video signals and generating violation alarm events and violation images based on the one or more entity objects includes: Extracting multidimensional feature values ​​from video image frames; One or more entity objects are identified based on the multidimensional feature values; Generate violation alarm events and violation images based on one or more entity objects.

10. The monitoring method according to claim 8, characterized in that, The method further includes the following steps: Using a scene recognition method based on FasterNet, the blasting dust scene is automatically distinguished. Analysis of the histogram of dust images in Lab space and elimination of histogram offsets are used to enhance the images of blasting dust. Based on the saturation estimation of the transmission map, a single image is quickly dehazed to eliminate the effects of blasting dust and restore a clear image. The YOLOv8s model is used to extract features from the enhanced image and identify the target objects in the image. Based on the extracted features of the target object, and combined with a pre-trained behavior classification model, the behavior of the target object is classified and identified.