Elevator safety guard method and system based on cloud side end fusion of multi-target detection

Abstract

The invention discloses an elevator safety guard method and system based on cloud side end fusion of multi-target detection, and relates to the field of target detection. Image data are collected in real time through a camera in an elevator, real-time reasoning is conducted through a trained target detection model, and electric vehicle and storage battery targets are detected; once illegal behaviors are detected, an alarm is given immediately, an elevator door is controlled to be kept open, and meanwhile the face detection and feature extraction process is started; face key point features are extracted through a dlib library, Euclidean distance matching is carried out on the face key point features and a preset resident face database, and identity recognition is achieved; a detection result and face feature data are uploaded to a cloud management platform, alarm information is generated and pushed to a property management APP, and property management personnel can remotely check the early warning accuracy and take corresponding measures. According to the method, the false alarm rate is effectively reduced, the recognition precision and the response speed are improved, and a complete treatment closed loop is formed.

Description

Technical Field

[0001] This invention relates to the field of target detection, and more specifically to an elevator safety protection method and system based on cloud-edge-device fusion of multi-target detection. Background Technology

[0002] In recent years, electric bicycles have been widely used due to their convenience, but the practice of users pushing them into elevators and charging them upstairs has brought about huge fire safety hazards.

[0003] The existing technical solutions have the following main shortcomings: 1. High false alarm rate: The sensor solution uses pressure sensors or laser beam sensors, which can only detect whether an object has entered, but cannot distinguish whether the object is an electric bicycle. It is very easy to misidentify strollers, bicycles, etc. as electric bicycles, resulting in false alarms and a poor user experience. 2. Slow response speed: The pure cloud-based analysis solution uploads all elevator monitoring video streams to the cloud server for analysis, which requires high network bandwidth, has a large analysis latency, cannot achieve real-time linkage control of elevator doors, and has a slow response speed. 3. Low recognition accuracy: In complex elevator environments (such as changes in lighting, occlusion, and different vehicle models), traditional image recognition algorithms have poor generalization ability, and insufficient accuracy and robustness. 4. Lack of closed-loop response: Most solutions only focus on detection and alarm, lacking a complete response process that includes linkage with the elevator control system, identification of unauthorized personnel, and intervention by property management personnel, thus failing to form an effective safety closed loop.

[0004] Therefore, there is an urgent need for an elevator safety protection method and system to overcome the problems existing in the current technology. Summary of the Invention

[0005] The purpose of this invention is to provide an elevator safety protection method and system based on cloud-edge-device fusion of multi-target detection, which features low false alarm rate, high recognition accuracy, fast response speed, and complete closed-loop handling, thereby preventing electric bicycles from entering elevators at the source and ultimately reducing the risk of building fires.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The elevator safety protection method based on multi-target detection and cloud-edge-device fusion includes the following steps: S1: Real-time image data inside the elevator is acquired via an in-elevator camera; S2: Load the trained target detection model on the edge computing device, perform real-time inference and prediction on the acquired image data, obtain the category, confidence level and target box coordinate information of each detected target, and determine whether there is electric vehicle entering the elevator based on this. S3: When an electric vehicle or battery target is detected, a local voice alarm is immediately triggered, and a control command is generated to keep the elevator door open, and the face detection and feature extraction process is started simultaneously. S4: Extract facial key point features from elevator videos using the dlib library, calculate the Euclidean distance between the feature and the key points of features in the preset resident face database, and select the feature with the smallest distance for identity matching; S5: Upload the detection data, including target detection results and facial features, to the cloud management platform via the MQTT protocol; S6: The cloud management platform generates alarm information based on the face matching results and pushes it to the property management APP. Property staff can verify the accuracy of the alarm through the APP. S7: The system continuously monitors the elevator status and automatically executes preset response rules based on the duration of unauthorized personnel's stay.

[0007] Furthermore, the target detection model in S2 is the YOLOv8n model, and the following optimization strategy is adopted during training: The elevator scene images were stitched together using Mosaic data augmentation technology. The activation probability was 0.8 in the early stage of training and dropped to 0.1 in the later stage of training. Adaptive image scaling technology is used to unify the input resolution; Introducing the ULSAM attention mechanism improves the accuracy of small target detection on batteries.

[0008] Furthermore, the process S4, prior to face matching, also includes: The detected faces are corrected by affine transformation to eliminate the influence of angular deviation on the Euclidean distance calculation.

[0009] Furthermore, S6 also includes: If the cumulative number of violations by the same resident exceeds the preset threshold, the cloud management platform will automatically trigger a manual intervention reminder.

[0010] Furthermore, the accuracy of the verification and early warning described in S6 specifically refers to: View elevator monitoring footage in real time via the app; If a misjudgment is confirmed, manually cancel the warning and restore elevator operation; If the warning is correct, a photo is taken as evidence and written into the database. At the same time, the person in the elevator is warned through the APP's voice call function.

[0011] Furthermore, the response rules described in S7 are as follows: If a person who violates the rules lingers for more than 30 seconds, a screenshot will be automatically taken and written to the log file. If you do not log out within 1 minute, send an emergency notification to the property management.

[0012] Another objective of this invention is to provide an elevator safety protection system based on cloud-edge-device fusion using multi-target detection. This system executes the aforementioned elevator safety protection method based on cloud-edge-device fusion using multi-target detection, including: Image acquisition module: The video output interface of the elevator camera is used to acquire monitoring images in real time; Target detection module: Deployed inside the edge computing unit, configured to run the YOLOv8n model, it detects electric vehicles, batteries and human faces in images in real time, and outputs target category, confidence score and coordinate information; Face recognition module: Integrated into the cloud management platform, configured to call the dlib library to extract facial key point features and perform Euclidean distance matching with the resident database; Edge computing unit: Deployed at the elevator site, it connects to the image acquisition module to acquire image data inside the elevator in real time, loads the target detection module to perform inference on the image, detects electric bicycles and batteries, and triggers a local warning and controls the elevator door to remain open when a target is detected; Cloud management platform: Receives and stores detection data from edge computing units, performs facial recognition and identification, manages violation event logs, pushes alarm information to user management terminals, and connects to the log storage database to execute detention timeout rules; The user management terminal communicates with the cloud management platform in real time, supporting the viewing of monitoring screens, verification of warnings, resolution of false judgments, and one-click linkage of personnel response.

[0013] Furthermore, the edge computing unit processes the target detection task through the TensorRT accelerated inference engine, achieving a real-time processing performance of ≥15 FPS at 1080P resolution; The cloud management platform uses an incremental training mechanism to continuously optimize the target detection model parameters using property verification feedback data and then sends them to the edge units for updates.

[0014] Another objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned elevator safety protection method based on multi-target detection and cloud-edge-device fusion.

[0015] Another objective of this invention is to provide an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor runs the computer program to enable the electronic device to perform the cloud-edge-device fusion elevator safety protection method based on multi-target detection.

[0016] In summary, the present invention has at least one of the following beneficial technical effects: 1. High recognition accuracy and low false alarm rate: It adopts a multi-object detection algorithm based on deep learning, which can accurately distinguish different objects such as electric bicycles, bicycles, and strollers, and significantly reduce the false alarm rate; 2. Fast real-time response: Edge computing mode places AI analysis locally, avoiding network transmission delays. The response time from detecting an electric vehicle to controlling the elevator door to open is extremely short (millisecond level), achieving true real-time interception. 3. Complete closed-loop management: It has formed a fully automated closed-loop management system of "detection-early warning-intervention-recording-tracing", which greatly improves the efficiency of property management and the level of security and prevention. 4. Assigning responsibility to individuals: By combining facial recognition technology, not only can vehicles be detected, but also the person pushing the cart can be identified, effectively tracing the violator and ensuring safety responsibility is fulfilled; 5. High system efficiency and stability: The cloud-edge-device converged architecture rationally allocates computing load, with the edge handling real-time inference and the cloud handling centralized management and big data analysis. Combined with containerization technology, it improves the system's scalability, maintainability, and security. Attached Figure Description

[0017] Figure 1 This is a flowchart of the present invention; Figure 2 This is a diagram distinguishing various vehicle models according to the present invention; Figure 3 This is a schematic diagram of the face recognition process of the present invention; Figure 4 This is a screenshot of the actual effect of the client app of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0019] This embodiment provides an elevator safety protection method based on multi-target detection and cloud-edge-device fusion, such as... Figure 1 As shown, it includes the following steps: S1: Real-time image data inside the elevator is acquired via an in-elevator camera; S2: Load the trained target detection model on the edge computing device, perform real-time inference and prediction on the acquired image data, obtain the category, confidence level and target box coordinate information of each detected target, and determine whether there is electric vehicle entering the elevator based on this. S3: When an electric vehicle or battery target is detected, a local voice alarm is immediately triggered, and a control command is generated to keep the elevator door open, and the face detection and feature extraction process is started simultaneously. S4: Extract facial key point features from elevator videos using the dlib library, calculate the Euclidean distance between the feature and the key points of features in the preset resident face database, and select the feature with the smallest distance for identity matching; S5: Upload the detection data, including target detection results and facial features, to the cloud management platform via the MQTT protocol; S6: The cloud management platform generates alarm information based on the face matching results and pushes it to the property management APP. Property staff can verify the accuracy of the alarm through the APP. S7: The system continuously monitors the elevator status and automatically executes preset response rules based on the duration of unauthorized personnel's stay.

[0020] The following is a detailed description with reference to specific embodiments.

[0021] S1: Real-time image data inside the elevator is obtained through an in-elevator camera.

[0022] A high-definition wide-angle camera and edge computing device (such as an NVIDIA Jetson NANO development board) are installed inside the elevator car. This device connects to the camera via USB or MIPI interface, to the elevator control system via serial port, and to the community LAN via Wi-Fi, ultimately connecting to the internet.

[0023] Image data from inside the elevator was acquired and the dataset was labeled. Labeling was performed using the YOLO bounding box format and normalized (x, y, w, h) coordinate representation. Differential images for various vehicle types are shown below. Figure 2 As shown: (1) Positive samples: motorcycles, electric vehicles, batteries, and frontal faces (including those wearing masks).

[0024] (2) Negative samples: bicycle, no frontal view (no need to label), carrying other objects and empty hands.

[0025] (3) 3-5 frontal photos of the face (used to extract facial recognition feature points) Use the LabelImg tool to annotate the bounding boxes in YOLO format, normalizing the coordinates (x, y, w, h). S2: Load the trained target detection model on the edge computing device, perform real-time inference and prediction on the acquired image data, obtain the category, confidence level and target box coordinate information of each detected target, and determine whether there is electric vehicle entering the elevator based on this.

[0026] The target detection model is the YOLOv8n model. The following optimization strategies were adopted during training: Mosaic data augmentation technology was used to stitch elevator scene images, with an activation probability of 0.8 in the early stage of training and reduced to 0.1 in the later stage of training; adaptive image scaling technology was used to unify the input resolution; and the ULSAM attention mechanism was introduced to improve the detection accuracy of small targets such as batteries.

[0027] Before inputting elevator monitoring footage into the YOLOv8n model, an innovative data preprocessing scheme was employed. Based on the characteristics of the YOLOv8n architecture, we implemented two key preprocessing techniques: First, we used Mosaic data augmentation, intelligently stitching together four elevator scene images to significantly improve the model's adaptability to complex elevator environments; second, we employed an adaptive image scaling strategy to automatically optimize the input image size, ensuring optimal detection performance across different elevator monitoring resolutions. It is important to note that while Mosaic augmentation effectively expands the diversity of training samples, YOLOv8n automatically reduces its usage frequency in the later stages of training to avoid potential performance degradation due to over-augmentation. This mechanism perfectly balances the advantages and disadvantages of data augmentation.

[0028] Later in the training process, Mosaic enhancements are disabled. This allows the model to focus on learning more refined features as training nears completion, resulting in better training performance. Adaptive image scaling is used to maintain the consistency of the input data. By uniformly scaling images of different sizes and resolutions to the standard size required for model training, the model is not limited by the original image size and can more effectively handle inputs of various sizes. This adaptive scaling not only ensures the model's processing power but also improves its generalization ability when processing images of different resolutions.

[0029] Through training with a single-stage object detection algorithm, its efficient depthwise separable convolution and C2f module structure can accurately learn the key features of electric vehicles and human faces, significantly improving detection and recognition accuracy. This network utilizes an improved backbone network and Feature Pyramid (PANet) to achieve multi-scale feature fusion, effectively solving common occlusion problems in elevator scenes (such as partial occlusion of electric vehicles) and accurately distinguishing easily confused objects (such as bicycles and electric vehicles).

[0030] Under complex lighting conditions, the algorithm's attention mechanism can adaptively focus on key areas, ensuring high accuracy in face recognition. Its anchor-free detector head and optimized loss function design enable the system to significantly reduce false alarm rates (such as misidentifying a regular stroller as an electric vehicle) and false negative rates while maintaining real-time performance.

[0031] Finally, the optimal model is loaded to predict the collected elevator interior image data. YOLOv8n is used to train the model and perform inference. The optimal model determines and obtains the category, confidence score, and corresponding bounding box coordinates of each detected target, thereby determining whether an electric vehicle has entered the elevator.

[0032] S3: When an electric vehicle or battery target is detected, a local voice alarm is immediately triggered, and a control command is generated to keep the elevator door open, and the face detection and feature extraction process is started simultaneously. This system employs a cascaded detection mechanism. First, a convolutional neural network model deployed on edge devices performs real-time analysis of the elevator monitoring video stream. Based on its optimized backbone network and feature pyramid structure, the model prioritizes electric vehicle detection. When an electric vehicle is detected entering the elevator, the system immediately triggers an early warning mechanism and simultaneously initiates the second-level facial recognition process: leveraging the network's lightweight nature, it quickly extracts passenger facial features within the same model framework and matches them against a community resident database for verification.

[0033] S4: Extract facial key point features from elevator videos using the dlib library, calculate the Euclidean distance between the feature and the key points of features in the preset resident face database, and select the feature with the smallest distance for identity matching.

[0034] like Figure 3 As shown, facial landmarks are extracted using methods such as dlib, and the coordinates of these landmarks are saved in a CSV file with the ID 'id'. The dlib library contains a large number of machine learning and graphics modeling algorithms and supports numerical algorithms. Using this library essentially involves calling the corresponding neural network structure and then passing pre-trained parameters to that network. It includes a face detector, a pre-trained facial landmark detector, and a face recognition model, providing a Python interface for easy use.

[0035] Before face matching, affine transformation correction is performed on the detected faces to eliminate the influence of angular deviation on Euclidean distance calculation.

[0036] S5: Upload the detection data, which includes the target detection results and facial features, to the cloud management platform via the MQTT protocol.

[0037] MQTT (Message Queuing Telemetry Transport) is a lightweight publish / subscribe messaging protocol designed specifically for the Internet of Things (IoT). It features: extremely low overhead (minimum message header is only 2 bytes, compared to kilobytes for HTTP); weak network adaptability (supporting high-latency / low-bandwidth networks, such as 4G signals in elevator shafts); and bidirectional communication (real-time bidirectional data streaming between devices and the cloud). Uploading to the cloud management platform via MQTT achieves an end-to-end latency of <200ms (meeting the real-time requirements of elevator door control). A single broker supports over 100,000 elevator terminals, maintaining a 99.9% message delivery rate even with a 10% packet loss rate in elevator shafts. Clustered brokers support access to millions of elevators in thousands of communities.

[0038] S6: The cloud management platform generates alarm information based on facial recognition results and pushes it to the property management app. Property staff verify the accuracy of the alarm through the app. If the cumulative number of violations by the same resident exceeds a preset threshold, the cloud management platform automatically triggers manual intervention. The preset threshold is set by the property staff. Verifying the accuracy of the alarm includes viewing the elevator monitoring screen in real time through the app; if a misjudgment is confirmed, the alarm is manually canceled and the elevator operation is restored; if the alarm is correct, photos are taken as evidence and written to the database, and the occupants in the elevator are warned via the app's voice call function.

[0039] S7: The system continuously monitors the elevator status and automatically executes preset response rules based on the duration of unauthorized personnel's stay. The specific response rules are as follows: if an unauthorized person stays for more than 30 seconds, a screenshot is automatically taken and written to the log file; if the person does not exit after more than 1 minute, an emergency notification is sent to the property management.

[0040] This embodiment provides an elevator safety protection system based on multi-target detection and cloud-edge-device fusion. The system executes the elevator safety protection method based on multi-target detection and cloud-edge-device fusion described in Embodiment 1, and includes: an image acquisition module, a target detection module, a face recognition module, an edge computing unit, a cloud management platform, and a user management terminal. Details are as follows.

[0041] Image acquisition module: The video output interface of the elevator camera is used to acquire monitoring images in real time; Target detection module: Deployed inside the edge computing unit, configured to run the YOLOv8n model, it detects electric vehicles, batteries and human faces in images in real time, and outputs target category, confidence score and coordinate information; Face recognition module: Integrated into the cloud management platform, configured to call the dlib library to extract facial key point features and perform Euclidean distance matching with the resident database; Edge computing unit: Deployed at the elevator site, it connects to the image acquisition module to acquire image data inside the elevator in real time, loads the target detection module to perform inference on the image, detects electric bicycles and batteries, and triggers a local warning and controls the elevator door to remain open when a target is detected; Cloud management platform: Receives and stores detection data from edge computing units, performs facial recognition and identification, manages violation event logs, pushes alarm information to user management terminals, and connects to the log storage database to execute detention timeout rules; User management terminal ( Figure 4 It communicates with the cloud management platform in real time, supporting the viewing of monitoring screens, verification of warnings, resolution of false judgments, and one-click linkage of personnel for response.

[0042] In addition, the edge computing unit uses TensorRT to accelerate the inference engine to process object detection tasks, achieving a real-time processing performance of ≥15 FPS at 1080P resolution; the cloud management platform uses an incremental training mechanism to continuously optimize the object detection model parameters using property verification feedback data and distributes them to the edge unit for updates.

[0043] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the cloud-edge-device fusion elevator safety protection method based on multi-target detection described in Embodiment 1.

[0044] This embodiment provides an electronic device, including a memory and a processor. The memory is used to store computer programs, and the processor runs the computer programs to enable the electronic device to execute the cloud-edge-device fusion elevator safety protection method based on multi-target detection described in Embodiment 1.

[0045] In summary, this invention employs a deep learning-based multi-object detection algorithm that accurately distinguishes between different objects such as electric bicycles, bicycles, and strollers, significantly reducing the false alarm rate and achieving high recognition accuracy with a low false alarm rate. The edge computing mode places AI analysis locally, avoiding network transmission latency. The response time from detecting an electric vehicle to controlling the elevator door opening is extremely short (milliseconds), achieving true real-time interception and rapid real-time response. It forms a fully automated closed-loop management system encompassing "detection-early warning-intervention-recording-tracing," greatly improving property management efficiency and security levels, with a complete closed-loop handling process. Combined with facial recognition technology, it can not only detect vehicles but also identify the person pushing the stroller, effectively tracing the violator and assigning responsibility for safety to specific individuals. The cloud-edge-device fusion architecture rationally allocates the computing load, with the edge handling real-time inference and the cloud handling centralized management and big data analysis. Combined with containerization technology, it improves the system's scalability, maintainability, and security, resulting in a highly efficient and stable system.

[0046] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0047] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0048] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0049] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0050] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A cloud-edge-device fusion method for elevator safety protection based on multi-target detection, characterized in that, Includes the following steps: S1: Real-time image data inside the elevator is acquired via an in-elevator camera; S2: Load the trained target detection model on the edge computing device, perform real-time inference and prediction on the acquired image data, obtain the category, confidence level and target box coordinate information of each detected target, and determine whether there is electric vehicle entering the elevator based on this. S3: When an electric vehicle or battery target is detected, a local voice alarm is immediately triggered, and a control command is generated to keep the elevator door open, and the face detection and feature extraction process is started simultaneously. S4: Extract facial key point features from elevator videos using the dlib library, calculate the Euclidean distance between the feature and the key points of features in the preset resident face database, and select the feature with the smallest distance for identity matching; S5: Upload the detection data, including target detection results and facial features, to the cloud management platform via the MQTT protocol; S6: The cloud management platform generates alarm information based on the face matching results and pushes it to the property management APP. Property staff can verify the accuracy of the alarm through the APP. S7: The system continuously monitors the elevator status and automatically executes preset response rules based on the duration of unauthorized personnel's stay.

2. The elevator safety protection method based on cloud-edge-device fusion using multi-target detection as described in claim 1, characterized in that, The target detection model in S2 is the YOLOv8n model, and the following optimization strategy is adopted during training: The elevator scene images were stitched together using Mosaic data augmentation technology. The activation probability was 0.8 in the early stage of training and dropped to 0.1 in the later stage of training. Adaptive image scaling technology is used to unify the input resolution; Introducing the ULSAM attention mechanism improves the accuracy of small target detection on batteries.

3. The elevator safety protection method based on cloud-edge-device fusion with multi-target detection according to claim 1, characterized in that, The S4 process, prior to face matching, also includes: The detected faces are corrected by affine transformation to eliminate the influence of angular deviation on the Euclidean distance calculation.

4. The elevator safety protection method based on cloud-edge-device fusion with multi-target detection according to claim 1, characterized in that, S6 further includes: If the cumulative number of violations by the same resident exceeds the preset threshold, the cloud management platform will automatically trigger a manual intervention reminder.

5. The elevator safety protection method based on cloud-edge-device fusion with multi-target detection according to claim 1, characterized in that, The accuracy of the verification and early warning described in S6 is specifically as follows: View elevator monitoring footage in real time via the app; If a misjudgment is confirmed, manually cancel the warning and restore elevator operation; If the warning is correct, a photo is taken as evidence and written into the database. At the same time, the person in the elevator is warned through the APP's voice call function.

6. The elevator safety protection method based on cloud-edge-device fusion with multi-target detection according to claim 1, characterized in that, The specific response rules described in S7 are as follows: If a person who violates the rules lingers for more than 30 seconds, a screenshot will be automatically taken and written to the log file. If you do not log out within 1 minute, send an emergency notification to the property management.

7. An elevator safety protection system based on multi-target detection and cloud-edge-device fusion, characterized in that, The elevator safety protection method based on multi-target detection and cloud-edge-device fusion, as described in any one of claims 1 to 6, includes: Image acquisition module: The video output interface of the elevator camera is used to acquire monitoring images in real time; Target detection module: Deployed inside the edge computing unit, configured to run the YOLOv8n model, it detects electric vehicles, batteries and human faces in images in real time, and outputs target category, confidence score and coordinate information; Face recognition module: Integrated into the cloud management platform, configured to call the dlib library to extract facial key point features and perform Euclidean distance matching with the resident database; Edge computing unit: Deployed at the elevator site, it connects to the image acquisition module to acquire image data inside the elevator in real time, loads the target detection module to perform inference on the image, detects electric bicycles and batteries, and triggers a local warning and controls the elevator door to remain open when a target is detected; Cloud management platform: Receives and stores detection data from edge computing units, performs facial recognition and identification, manages violation event logs, pushes alarm information to user management terminals, and connects to the log storage database to execute detention timeout rules; The user management terminal communicates with the cloud management platform in real time, supporting the viewing of monitoring screens, verification of warnings, resolution of false judgments, and one-click linkage of personnel response.

8. The elevator safety protection system based on multi-target detection and cloud-edge-device fusion according to claim 7, characterized in that: The edge computing unit processes target detection tasks through the TensorRT accelerated inference engine, achieving a real-time processing performance of ≥15 FPS at 1080P resolution. The cloud management platform uses an incremental training mechanism to continuously optimize the target detection model parameters using property verification feedback data and then sends them to the edge units for updates.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the cloud-edge-device fusion elevator safety protection method based on multi-target detection as described in any one of claims 1 to 6.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the cloud-edge-device fusion elevator safety protection method based on multi-target detection as described in any one of claims 1 to 6.

Images