Low-voltage switch automatic management method based on RFID and image AI identification technology

Abstract

The invention relates to a low-voltage switch automatic management method based on RFID and an image AI identification technology, and the method comprises the following steps: S1, installing an anti-metal RFID tag for each piece of equipment, and setting a camera; s2, establishing a device-account-physical label ternary associated digital asset ledger; s3, an LANDMARC improved algorithm is adopted, high-precision three-dimensional positioning is achieved through eight reference tags, meanwhile, personnel identities and approaching target equipment information are automatically read, personnel and equipment identity data are positioned in real time, and the data are transmitted to a monitoring system; s4, the camera captures a video stream in the whole process of operating the switch by the personnel, and analyzes the current state of the switch and the personnel action; s5, performing cross verification according to the RFID three-dimensional positioning result, the current state of the switch and the personnel action recognition result; and S6, according to the operation and alarm whole process log data, binding RFID and image data, and generating a standardized event record. The equipment operation safety level and the management efficiency are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of equipment management, and in particular to a low-voltage switch automation management method based on RFID and image AI recognition technology. Background Art

[0002] In recent years, with the development of intelligent power distribution technology, the demand for automated management of low-voltage switches and distribution equipment has continued to increase. Traditional low-voltage switch management methods largely rely on manual operation and paper or manual records. This poses challenges such as difficulty in uniquely confirming personnel identities, difficulty in fully monitoring operational behavior, and slow response to abnormal risks. In actual operation, safety hazards such as misoperation, unauthorized operation, omissions in records, and difficulty in accountability have become prominent shortcomings in power distribution management.

[0003] At the same time, the increasing variety and quantity of devices, their widespread distribution, and the demand for sophisticated management make it difficult for a single technology to meet the management needs of both high security and high efficiency. Traditional video surveillance can only provide passive backtracking, lacking deep linkage and verification of the authenticity of operator identities and actions. While single RFID technology can provide identity recognition and fixed-point positioning capabilities, it has limited control over complex dynamic operations. Summary of the Invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a low-voltage switch automation management method based on RFID and image AI recognition technology to effectively improve the equipment operation safety level and management efficiency.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A low-voltage switch automation management method based on RFID and image AI recognition technology includes the following steps:

[0007] S1: Install anti-metal RFID tags on each low-voltage switch and distribution equipment. Bind the tag ID to the physical coordinates of the equipment and deploy eight reference tags in the operation area. Set up cameras and calibrate the viewing angle to cover the operation panel and personnel activity area.

[0008] S2: Obtain a list of low-voltage switches and power distribution equipment, system account and permission allocation plan, bind a unique RFID tag to each switch and device, enter asset information, and establish a digital asset ledger with a three-way association of device, account, and physical tag;

[0009] S3: When entering the operation area, personnel carry their personal RFID cards and use the improved LANDMARC algorithm to achieve high-precision three-dimensional positioning using 8 reference tags. The algorithm automatically reads the personnel identity and the target equipment information they are approaching, and transmits the personnel and equipment identity data to the monitoring system in real time.

[0010] S4: The camera captures the video stream of the entire process of the person operating the switch and analyzes the current state of the switch and the person's actions;

[0011] S5: Based on the RFID three-dimensional positioning results, the current state of the switch and the results of personnel action recognition, cross-checking is performed to ensure the uniqueness and authenticity of the operation record. If an illegal operation is detected, the system will automatically alarm in real time, link the access control or provide on-site voice prompts;

[0012] S6: Based on the operation and alarm process log data, bind RFID and image data to generate standardized event records.

[0013] Furthermore, S1 is specifically:

[0014] Install anti-metal RFID tags on each low-voltage switch and power distribution equipment, register the tag ID with the equipment number and its spatial coordinates in the background asset management system, and form a spatial distribution map of the equipment;

[0015] Eight reference tags are distributed at the four horizontal corners of the operating area, the ground, and the top corners according to the principle of the eight vertices of a cube to ensure the formation of a three-dimensional positioning coordinate system. The dimensions are measured using the operating area as the x, y, and z axes, and the eight corners are numbered Ref1 to Ref8.

[0016] The industrial camera is installed in a top-down or oblique perspective, covering the operation panel and the main personnel activity area, and the calibration plate is used to calibrate the camera's internal and external parameters.

[0017] Furthermore, the improved LANDMARC algorithm is used to achieve high-precision 3D positioning using 8 reference tags, as follows:

[0018] Based on 8 anti-metal RFID reference tags, a three-dimensional positioning reference grid is formed;

[0019] Define the person carrying a personal RFID card as the target tag, and the i-th reference tag Ref i The RSSI difference ΔRSSI is:

[0020] ΔRSSI i =RSSI target -RSSI ref,i ;

[0021] Among them, RSSI target Target tag received signal strength, RSSI ref,i is the signal strength of the i-th reference tag; ΔRSSI i Converted into the three-dimensional space distance weight between the target label and the reference label

[0022] d i=α·ΔRSSI i +β;

[0023] Among them, α, β are environmental attenuation coefficients;

[0024] Select the top k reference labels with the smallest signal difference from the target label and calculate the 3D weighted coordinates:

[0025]

[0026] Among them, w j is the weight of the j-th reference label;

[0027] The Kalman filter is introduced to smooth the positioning results. The state at the previous moment and the observation at the current moment are used to dynamically correct and optimize the positioning results to suppress the errors caused by signal fluctuations.

[0028] Furthermore, Kalman filtering is introduced to smooth the positioning results, as follows:

[0029] Assume that the position estimate of the target label at each moment is X t =[x t ,y t ,z t ] T ; Wherein, the superscript T represents transposition;

[0030] Based on the last status Predict current location

[0031]

[0032] Where F is the state transfer matrix;

[0033] At the same time, update the prior covariance:

[0034] P t|t-1 =FP t-1|t-1 F T +Q;

[0035] Among them, P t|t-1 is the state prediction covariance; Q is the process noise covariance matrix;

[0036] Collect the positioning observation value at this moment:

[0037] Z t =H·X t +v t ;

[0038] Where Zt is the observation value; vt is the observation noise; H is the observation matrix;

[0039] Calculate the Kalman gain:

[0040] K t =P t∣t-1 H T (HP t∣t-1 H T +R) -1 ;

[0041] Among them, K t is the Kalman gain, which is used to suppress positioning jumps caused by signal jitter; R is the observation noise covariance matrix;

[0042] Use observations to modify the predicted state:

[0043]

[0044] in, is the current best estimate;

[0045] Update the covariance:

[0046] P t|t =(IK t H)P t|t-1 ;

[0047] Where I is the identity matrix;

[0048] Each time you locate and sample, follow the above steps and output This is the smoothed high-confidence three-dimensional position.

[0049] Furthermore, the camera captures the video stream of the entire process of the person operating the switch, and analyzes the current state of the switch and the person's actions, specifically:

[0050] The camera captures the video stream of the entire process of the person operating the switch and performs preprocessing, including image denoising, normalization, illumination compensation, dynamic background modeling, and ROI block extraction. Only the relevant images of the switch and the person's hand are retained and output as the ROI image stream;

[0051] Based on the ROI image stream, the target detection model captures the switch area in each frame of the video and outputs the switch status category. If there is an indicator light, the status is determined by combining color analysis.

[0052] Use OpenPose to identify the coordinates of a person's fingers, and analyze the continuous motion trajectory of the hand based on a time-series convolutional neural network to achieve the division of operation start, execution, and end;

[0053] Frame-level linkage RFID three-dimensional positioning determines the correspondence between people in the picture and target devices, and finally outputs the recognition results.

[0054] Further preprocessing is as follows:

[0055] A Gaussian filter is first applied to the raw image captured for each frame of the video stream to remove photosensitive noise and smooth textures. The denoising results are then normalized, histogram equalized, or gamma corrected to eliminate the effects of lighting and device differences and improve recognition robustness.

[0056] The background modeling method is used to obtain the foreground binary mask, effectively separating the person from the device operation area and the permanent static background;

[0057] Combined with deep learning target detection, ROIs of the hand and switch areas are extracted, the original large image is cropped, and the final ROI image stream is output.

[0058] Furthermore, the background modeling method is used to obtain the foreground binary mask to effectively separate the person from the device operation area and the permanent static background, as follows:

[0059] For pixel p, modeled by K Gaussian distributions, is the pixel value at time t:

[0060]

[0061] in, is the weight of the i′th Gaussian distribution; is the mean of the i′th Gaussian distribution; is the variance of the i′th Gaussian distribution; is the probability density function of the normal distribution;

[0062] Normal distribution probability density expression:

[0063]

[0064] For the current pixel Compare with each component j':

[0065]

[0066] If the match is established, write down all the matching components and press Sort from largest to smallest, the top B are identified as background models. If the current pixel matches one of the top B components, it is considered the background, otherwise it is the foreground.

[0067] For the matched components:

[0068]

[0069] For unmatched components, their weights are only exponentially decayed:

[0070]

[0071] Among them, ρ is the learning rate; α is the weight update rate;

[0072] Perform morphological operations on the mask to eliminate small noises and holes and obtain a coherent operation area.

[0073] Furthermore, OpenPose is used to identify the coordinates of the person's fingers. Based on the time series convolutional neural network, the continuous motion trajectory of the hand is analyzed to realize the division of the operation start, execution, and end, as follows:

[0074] Lock the ROI area based on target detection and intercept the image stream;

[0075] Use OpenPose to extract the finger key points P in each frame ROI t (21):

[0076]

[0077] in, is the pixel coordinate of the k′th key point at time t;

[0078] Normalize the key points of each frame:

[0079]

[0080] in, is the pixel coordinate of the wrist key point in the current frame; w ROI ,h ROI The width and height of the current ROI area; is the normalized coordinate of the k′th key point;

[0081] Construct dynamic feature sequences, including speed acceleration and the global motion energy E t :

[0082]

[0083] Construct a T-frame sliding window to input the TCN model and output the start, execution, and end stage labels corresponding to each frame.

[0084] The low-voltage switch automation management system based on RFID and image AI recognition technology includes a processor, a memory, and a computer program stored on the memory. When the processor executes the computer program, it specifically executes the steps in the low-voltage switch automation management method based on RFID and image AI recognition technology as described above.

[0085] A computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the above method steps.

[0086] The present invention has the following beneficial effects:

[0087] 1. This invention uses anti-metal RFID tags combined with eight reference tags in an optimized layout to achieve three-dimensional spatial positioning based on the improved LANDMARC algorithm. Each low-voltage switch, power distribution equipment, and operator has a unique RFID identifier. The system automatically establishes a three-element digital asset ledger of equipment, account number, and physical tag, achieving precise positioning of equipment and operators and digital, real-time visual management of asset information, greatly improving the intelligent level of on-site operation and maintenance management.

[0088] 2. This invention collects images and video streams of the entire operation process in real time, combines RFID precise three-dimensional positioning and identity recognition data, and simultaneously uses image AI to perform intelligent analysis of switch status and personnel actions. Multi-source data fusion and cross-verification ensure the uniqueness, authenticity and traceability of operation records.

[0089] 3. The present invention automatically generates a standardized operation event log, including normal operations and all alarm data, so that the entire process of operation events can be traced and verified. Once unauthorized switches, dangerous behaviors or illegal operations are found, the system can automatically link access control, voice, alarm and other security facilities to respond and issue warnings in the first time, thereby improving operational safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0091] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0092] refer to Figure 1 In this embodiment, a low-voltage switch automation management method based on RFID and image AI recognition technology is provided, which is characterized by comprising the following steps:

[0093] S1: Install anti-metal RFID tags on each low-voltage switch and distribution equipment. Bind the tag ID to the physical coordinates of the equipment. Deploy eight reference tags in the operating area (optimize the layout to a cube vertex distribution to improve the LANDMARC algorithm's three-dimensional positioning accuracy to ±10cm). Set up a camera (2 megapixels, 30fps, with infrared fill light), and calibrate the viewing angle to cover the operating panel and the personnel activity area.

[0094] S2: Obtain a list of low-voltage switches and power distribution equipment, system account and permission allocation plan, bind a unique RFID tag to each switch and device, enter asset information, and establish a digital asset ledger with a three-way association of device, account, and physical tag;

[0095] The power distribution equipment includes HPLC modular equipment, electric energy metering equipment and capacitor equipment;

[0096] S3: When entering the operation area, personnel carry their personal RFID cards and use the improved LANDMARC algorithm to achieve high-precision three-dimensional positioning using 8 reference tags. The algorithm automatically reads the personnel identity and the target equipment information they are approaching, and transmits the personnel and equipment identity data to the monitoring system in real time.

[0097] S4: The camera captures the video stream of the entire process of the person operating the switch and analyzes the current state of the switch and the person's actions;

[0098] S5: Based on the RFID three-dimensional positioning results, the current state of the switch and the results of the personnel action recognition, a cross-check is performed to ensure the uniqueness and authenticity of the operation record. If an illegal operation is detected (such as unauthorized switch, dangerous action), the system will automatically alarm in real time, link the access control or provide on-site voice prompts;

[0099] S6: Based on the operation and alarm process log data, bind RFID and image data to generate standardized event records.

[0100] In this embodiment, S1 is specifically:

[0101] Install anti-metal RFID tags on each low-voltage switch and power distribution equipment, register the tag ID with the equipment number and its spatial coordinates in the background asset management system, and form a spatial distribution map of the equipment;

[0102] Eight reference tags (RFID passive tags) are distributed at the four horizontal corners of the operating area and the four corners on the ground and above, following the principle of the eight vertices of a cube, to ensure the formation of a three-dimensional positioning coordinate system. The dimensions of the operating area are measured along the x, y, and z axes, and the eight corners are numbered Ref1 to Ref8. The specific coordinates of each point (which can be accurately measured using a laser rangefinder) are convenient for reference node signal strength fingerprint calibration in the LANDMARC algorithm.

[0103] Industrial cameras are installed in a top-down or oblique perspective, covering the operation panel and the main personnel activity areas. A calibration board (checkerboard board) is used to calibrate the internal and external parameters of the camera to ensure that subsequent image AI can be accurately associated with the physical coordinates of the space, realizing the spatial traceability of human motion and equipment status analysis.

[0104] In this embodiment, an improved LANDMARC algorithm is used to achieve high-precision three-dimensional positioning using 8 reference tags, as follows:

[0105] Based on 8 anti-metal RFID reference tags, a three-dimensional positioning reference grid is formed;

[0106] Define the person carrying a personal RFID card as the target tag, and the i-th reference tag Ref i The RSSI difference ΔRSSI is:

[0107] ΔRSSI i =RSSI target -RSSI ref,i ;

[0108] Among them, RSSI target Target tag received signal strength, RSSI ref,i is the signal strength of the i-th reference tag; ΔRSSI i Converted into the three-dimensional space distance weight between the target label and the reference label

[0109] d i =α·ΔRSSI i +β;

[0110] Among them, α, β are environmental attenuation coefficients;

[0111] Select the top k reference labels (usually k = 4) with the smallest signal difference from the target label and calculate the 3D weighted coordinates:

[0112]

[0113] Among them, w j is the weight of the j-th reference label;

[0114] The Kalman filter is introduced to smooth the positioning results. The state at the previous moment and the observation at the current moment are used to dynamically correct and optimize the positioning results to suppress the errors caused by signal fluctuations.

[0115] In this embodiment, a Kalman filter is introduced to smooth the positioning results, as follows:

[0116] Assume that the position estimate of the target label at each moment is X t =[x t ,y t ,z t ] T ; Wherein, the superscript T represents transposition;

[0117] Based on the last status Predict current location

[0118]

[0119] Where F is the state transfer matrix;

[0120] At the same time, update the prior covariance:

[0121] Pt|t-1 =FP t-1|t-1 F T +Q;

[0122] Among them, P t|t-1 is the state prediction covariance; Q is the process noise covariance matrix;

[0123] Collect the positioning observation value at this moment:

[0124] Z t =H·X t +v t ;

[0125] Where Zt is the observation value; vt is the observation noise; H is the observation matrix;

[0126] Calculate the Kalman gain:

[0127] K t =P t∣t-1 H T (HP t∣t-1 H T +R) -1 ;

[0128] Among them, K t is the Kalman gain, which is used to suppress positioning jumps caused by signal jitter; R is the observation noise covariance matrix;

[0129] Use observations to modify the predicted state:

[0130]

[0131] in, is the current optimal estimate (smoothed output);

[0132] Update the covariance:

[0133] P t|t =(IK t H)P t|t-1 ;

[0134] Where I is the identity matrix;

[0135] Each time you locate and sample, follow the above steps and output This is the smoothed high-confidence three-dimensional position.

[0136] In this embodiment, the camera captures the video stream of the entire process of a person operating a switch, and analyzes the current state of the switch and the person's actions, specifically:

[0137] The camera captures the video stream of the entire process of the person operating the switch and preprocesses it, including image denoising, normalization, lighting compensation, dynamic background modeling (such as using MOG), and ROI block extraction. Only the relevant images of the switch and the person's hand are retained, and the ROI image stream is output;

[0138] Based on the ROI image stream, the switch area is captured for each frame of the video using an object detection model (YOLO, Faster R-CNN) and the switch status category (e.g., "closed" / "open") is output. If an indicator light is present, the status is determined using color analysis (HSV space).

[0139] Use OpenPose to identify the coordinates of a person's fingers, and analyze the continuous motion trajectory of the hand based on a time-series convolutional neural network to achieve the division of operation start, execution, and end;

[0140] Frame-level linkage RFID three-dimensional positioning determines the correspondence between people in the picture and target devices, and finally outputs the recognition results.

[0141] In this embodiment, the preprocessing is as follows:

[0142] A Gaussian filter is first applied to the raw image captured for each frame of the video stream to remove photosensitive noise and smooth textures. The denoising results are then normalized, histogram equalized, or gamma corrected to eliminate the effects of lighting and device differences and improve recognition robustness.

[0143] The background modeling method is used to obtain the foreground binary mask, effectively separating the person from the device operation area and the permanent static background;

[0144] Combined with deep learning target detection, ROIs of the hand and switch areas are extracted, the original large image is cropped, and the final output ROI image stream provides accurate and interference-free input for subsequent AI recognition of switch status and hand movements.

[0145] In this embodiment, a background modeling method is used to obtain a foreground binary mask to effectively separate the person from the device operation area and the permanent static background, as follows:

[0146] For pixel p, modeled by K Gaussian distributions, is the pixel value at time t (can be grayscale or RGB vector):

[0147]

[0148] in, is the weight of the i′th Gaussian distribution; is the mean of the i′th Gaussian distribution; is the variance of the i′th Gaussian distribution; is the probability density function of the normal distribution;

[0149] Normal distribution probability density expression:

[0150]

[0151] For the current pixel Compare with each component j':

[0152]

[0153] If the match is established, write down all the matching components and press Sort from largest to smallest, the top B are identified as background models. If the current pixel matches one of the top B components, it is considered the background, otherwise it is the foreground.

[0154] For the matched components:

[0155]

[0156] For unmatched components, their weights are only exponentially decayed:

[0157]

[0158] Among them, ρ is the learning rate; α is the weight update rate;

[0159] Perform morphological operations (such as opening and closing operations) on the mask to eliminate small noise and holes and obtain a coherent operation area.

[0160] In this embodiment, OpenPose is used to identify the coordinates of a person's fingers, and a temporal convolutional neural network is used to analyze the continuous motion trajectory of the hand to achieve the division of operation start, execution, and end, as follows:

[0161] Lock the ROI area based on target detection and intercept the image stream;

[0162] Use OpenPose to extract the finger key points P in each frame ROI t (21):

[0163]

[0164] in, is the pixel coordinate of the k′th key point at time t;

[0165] Normalize the key points of each frame:

[0166]

[0167] in, is the pixel coordinate of the wrist key point in the current frame; w ROI , hROI The width and height of the current ROI area; is the normalized coordinate of the k′th key point;

[0168] Construct dynamic feature sequences, including speed acceleration and the global motion energy E t :

[0169]

[0170] Construct a T-frame sliding window to input the TCN model and output the start, execution, and end stage labels corresponding to each frame.

[0171] In this embodiment, a low-voltage switch automation management system based on RFID and image AI recognition technology is also provided, which is characterized in that it includes a processor, a memory, and a computer program stored on the memory. When the processor executes the computer program, it specifically executes the steps in the low-voltage switch automation management method based on RFID and image AI recognition technology as described above.

[0172] In this embodiment, a computer storage medium is also provided, wherein the computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the above method steps.

[0173] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0174] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0175] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0176] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0177] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above-disclosed technical content to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention remain within the scope of protection of the present invention.

Claims

1. The low-voltage switch automation management method based on RFID and image AI recognition technology is characterized by: The following steps are involved: S1: Install anti-metal RFID tags on each low-voltage switch and distribution equipment. The tag ID is bound to the physical coordinates of the equipment, and 8 reference tags are deployed in the operating area. Set up the camera and calibrate the viewing angle to cover the operation panel and personnel activity area; S2: Obtain a list of low-voltage switches and power distribution equipment, system account and permission allocation plan, bind a unique RFID tag to each switch and device, enter asset information, and establish a digital asset ledger with a three-way association of device, account, and physical tag; S3: When entering the operation area, personnel carry their personal RFID cards and use the improved LANDMARC algorithm to achieve high-precision three-dimensional positioning using 8 reference tags. The algorithm automatically reads the personnel identity and the target equipment information they are approaching, and transmits the personnel and equipment identity data to the monitoring system in real time. S4: The camera captures the video stream of the entire process of the person operating the switch and analyzes the current state of the switch and the person's actions; S5: Based on the RFID three-dimensional positioning results, the current state of the switch and the results of personnel action recognition, cross-checking is performed to ensure the uniqueness and authenticity of the operation record. If an illegal operation is detected, the system will automatically alarm in real time, link the access control or provide on-site voice prompts; S6: Based on the operation and alarm process log data, bind RFID and image data to generate standardized event records.

2. The low-voltage switch automation management method based on RFID and image AI recognition technology according to claim 1 is characterized in that: The S1 is specifically: Install anti-metal RFID tags on each low-voltage switch and power distribution equipment, register the tag ID with the equipment number and its spatial coordinates in the background asset management system, and form a spatial distribution map of the equipment; Eight reference tags are distributed at the four horizontal corners of the operating area, the ground, and the top corners according to the principle of the eight vertices of a cube to ensure the formation of a three-dimensional positioning coordinate system. The dimensions are measured using the operating area as the x, y, and z axes, and the eight corners are numbered Ref1 to Ref8. The industrial camera is installed in a top-down or oblique perspective, covering the operation panel and the main personnel activity area, and the calibration plate is used to calibrate the camera's internal and external parameters.

3. The low-voltage switch automation management method based on RFID and image AI recognition technology according to claim 1 is characterized in that: The improved LANDMARC algorithm is used to achieve high-precision three-dimensional positioning using 8 reference tags, as follows: Based on 8 anti-metal RFID reference tags, a three-dimensional positioning reference grid is formed; Define the person carrying a personal RFID card as the target tag, and the i-th reference tag Ref i The RSSI difference ΔRSSI is: ΔRSSI i =RSSI target -RSSI ref,i ; Among them, RSSI target Target tag received signal strength, RSSI ref,i is the signal strength of the i-th reference tag; ΔRSSI i Converted into the three-dimensional space distance weight between the target label and the reference label d i =α·ΔRSSI i +b; Among them, α, β are environmental attenuation coefficients; Select the top k reference labels with the smallest signal difference from the target label and calculate the 3D weighted coordinates: Among them, w j is the weight of the j-th reference label; The Kalman filter is introduced to smooth the positioning results. The state at the previous moment and the observation at the current moment are used to dynamically correct and optimize the positioning results to suppress the errors caused by signal fluctuations.

4. The low-voltage switch automation management method based on RFID and image AI recognition technology according to claim 3 is characterized in that: The Kalman filter is introduced to smooth the positioning results, as follows: Assume that the position estimate of the target label at each moment is X t =[x t ,y t ,z t ] T ; Wherein, the superscript T represents transposition; based on the previous state Predict current location Where F is the state transfer matrix; At the same time, update the prior covariance: P t|t-1 =FP t-1|tt-1 F T +Q; Among them, P t|t-1 is the state prediction covariance; Q is the process noise covariance matrix; Collect the positioning observation value at this moment: Z t =H·X t +v t ; Where Zt is the observation value; vt is the observation noise; H is the observation matrix; Calculate the Kalman gain: K t =P t∣t-1 H T (HP t∣t-1 H T +R) -1 ; Among them, K t is the Kalman gain, which is used to suppress positioning jumps caused by signal jitter; R is the observation noise covariance matrix; Use observations to modify the predicted state: in, is the current best estimate; Update the covariance: P t|t =(I-K t H)P t|t-1 ; Where I is the identity matrix; Each time you locate and sample, follow the above steps and output This is the smoothed high-confidence three-dimensional position.

5. The low-voltage switch automation management method based on RFID and image AI recognition technology according to claim 1 is characterized in that: The camera captures the video stream of the entire process of the person operating the switch and analyzes the current state of the switch and the person's actions, specifically: The camera captures the video stream of the entire process of the person operating the switch and performs preprocessing, including image denoising, normalization, illumination compensation, dynamic background modeling, and ROI block extraction. Only the relevant images of the switch and the person's hand are retained and output as the ROI image stream; Based on the ROI image stream, the target detection model captures the switch area in each frame of the video and outputs the switch status category. If there is an indicator light, the status is determined by combining color analysis. Use OpenPose to identify the coordinates of a person's fingers, analyze the continuous motion trajectory of the hand based on a time-series convolutional neural network, and divide the operation into start, execution, and end points; Frame-level linkage RFID three-dimensional positioning determines the correspondence between people in the picture and target devices, and finally outputs the recognition results.

6. The low-voltage switch automation management method based on RFID and image AI recognition technology according to claim 5 is characterized in that: The preprocessing is as follows: A Gaussian filter is first applied to the raw image captured for each frame of the video stream to remove photosensitive noise and smooth textures. The denoising results are then normalized, histogram equalized, or gamma corrected to eliminate the effects of lighting and device differences and improve recognition robustness. The background modeling method is used to obtain the foreground binary mask, effectively separating the person from the device operation area and the permanent static background; Combined with deep learning target detection, ROIs of the hand and switch areas are extracted, the original large image is cropped, and the final ROI image stream is output.

7. The low-voltage switch automation management method based on RFID and image AI recognition technology according to claim 6 is characterized in that: The background modeling method is used to obtain a foreground binary mask, which effectively separates the person from the device operation area and the resident static background, as follows: For pixel p, modeled by K Gaussian distributions, is the pixel value at time t: in, is the weight of the i′th Gaussian distribution; is the mean of the i′th Gaussian distribution; is the variance of the i′th Gaussian distribution; is the probability density function of the normal distribution; Normal distribution probability density expression: For the current pixel Compare with each component j': If the match is established, write down all the matching components and press Sort from largest to smallest, the top B are identified as background models. If the current pixel matches one of the top B components, it is considered the background, otherwise it is the foreground. For the matched components: For unmatched components, their weights are only exponentially decayed: Among them, ρ is the learning rate; α is the weight update rate; Perform morphological operations on the mask to eliminate small noises and holes and obtain a coherent operation area.

8. The low-voltage switch automation management method based on RFID and image AI recognition technology according to claim 6 is characterized in that: The method uses OpenPose to identify the coordinates of a person's fingers and analyzes the continuous motion trajectory of the hand based on a temporal convolutional neural network to achieve the division of the operation start, execution, and end. Specifically, the method locks the ROI area based on target detection and intercepts the image stream. Use OpenPose to extract the finger key points P in each frame ROI t (21): in, is the pixel coordinate of the k′th key point at time t; Normalize the key points of each frame: in, is the pixel coordinate of the wrist key point in the current frame; w ROI , h ROI The width and height of the current ROI area; is the normalized coordinate of the k′th key point; Construct dynamic feature sequences, including speed acceleration and the global motion energy E t : Construct a T-frame sliding window to input the TCN model and output the start, execution, and end stage labels corresponding to each frame.

9. Low-voltage switch automation management system based on RFID and image AI recognition technology, characterized by: It includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the low-voltage switch automation management method based on RFID and image AI recognition technology as described in any one of claims 1 to 8.

10. A computer storage medium, characterized in that The computer storage medium stores a plurality of instructions, which are suitable for being loaded by a processor and executing the method steps according to any one of claims 1 to 8.

Images