Fume hood window height intelligent monitoring and alarming system and method based on visual target detection

Abstract

The invention provides a ventilation cabinet window height intelligent monitoring and alarm system and method based on visual target detection, and the system collects a window image through a camera in a non-contact manner, and precisely locates the position of a window through a deep learning model based on a direction bounding box. And boundary crossing judgment is carried out by calculating the position relation between the key points and the preset safety line. The system innovatively introduces a delay alarm and multi-level early warning mechanism, effectively reduces the false alarm rate, and has the functions of graphical calibration and system self-diagnosis. According to the invention, the problems of complex deployment and easy damage of the existing contact type sensor are solved, intelligent and high-reliability real-time monitoring of the safety state of the ventilation cabinet is realized, and the method is suitable for safety management of various laboratories.

Description

Technical Field

[0001] This invention relates to the field of laboratory safety monitoring technology, and in particular to an intelligent monitoring and alarm system and method for the height of fume hood windows based on visual target detection. Background Technology

[0002] Fume hoods are essential protective equipment in laboratories to prevent the leakage of harmful gases and ensure the health of operators. Their performance is closely related to the opening height of the viewing window—an excessively high opening height disrupts the negative pressure airflow inside the hood, causing harmful gases to escape and increasing occupational exposure risks; while an excessively low opening height restricts the experimental operating space. Therefore, international standards (such as ASHRAE 110) typically stipulate that the safe opening height of the viewing window should not exceed 40 cm or should be dynamically adjusted according to face velocity.

[0003] Current automated monitoring solutions mostly involve installing contact sensors (such as ultrasonic, infrared photocells, or magnetic sensors) on the fume hood rails. These solutions have significant drawbacks: they require modifications to the fume hood structure, potentially affecting its original airtightness and safety certification; the sensors are susceptible to chemical corrosion and dust contamination, leading to malfunctions; and the mechanical installation is complex, resulting in high maintenance costs.

[0004] Therefore, there is an urgent need in this field for a non-contact, high-precision, high-reliability, and easy-to-deploy intelligent monitoring solution for the height of fume hood windows. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent monitoring and alarm system and method for the height of fume hood windows based on visual target detection.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A visual target detection-based intelligent monitoring and alarm system for the height of fume hood windows, including... An image acquisition module is used to continuously acquire video streams containing the viewing area of ​​the fume hood; the image acquisition module is an industrial camera fixed in front of the fume hood for acquiring video streams.

[0007] An image processing module, connected to the image acquisition module, is used to preprocess the video stream images and identify the window position based on a deep learning object detection model; the image processing module is mounted on an edge computing device and runs an optimized deep learning object detection model (such as YOLOv8-OBB).

[0008] A safety monitoring module, connected to the image processing module, is used to make boundary judgments based on the identified window position and the preset safety height line. An alarm module, connected to the safety monitoring module, is used to trigger a multi-mode alarm signal when the safety monitoring module determines that a continuous boundary crossing has occurred. The alarm module makes boundary crossing judgments based on the upper edge vertex of the directional bounding box and introduces a delay mechanism to reduce false alarms.

[0009] The system calibration module is used to establish the mapping relationship between the physical world safety height and the safety line in the image coordinate system.

[0010] Preferably, the deep learning target detection model used by the image processing module is an output model based on oriented bounding boxes, which is used to detect the bottom beam or handle of the viewport and output the coordinates of its bounding box vertices with rotation angle.

[0011] Preferably, the boundary judgment logic of the security monitoring module is as follows: using the coordinates of the upper edge vertex of the directional bounding box, the cross product method is used to determine whether the vertex is located above the preset safety line.

[0012] Preferably, the security monitoring module further includes a delay judgment unit, which, after detecting an out-of-bounds state, determines it as a valid violation and triggers the alarm module only if the state continues to exceed a preset delay threshold.

[0013] Preferably, the alarm module supports multi-level alarms, including: Level 1 warning: A visual alert is triggered when the viewing window height approaches the safety line. Level 2 alarm triggers sound, light, voice, and remote notifications when the viewing window height continuously exceeds the safety line.

[0014] Preferably, it also includes a system self-diagnosis module, which is used to monitor the working status and image quality of the image acquisition module, and trigger a system fault alarm when an abnormality occurs.

[0015] A method for intelligent monitoring and alarm of fume hood window height based on visual target detection, the method comprising the following steps: S1: The system calibration module completes the calibration from the physical safety height to the image safety line; S2: The image acquisition module acquires video images of the fume hood viewing area in real time. S3: The image processing module preprocesses the image and uses a deep learning model to identify the precise location of the window; S4: The security monitoring module compares the identified window position with the image security line to determine whether an out-of-bounds violation has occurred. If a continuous out-of-bounds violation occurs, an alarm signal will be triggered via the alarm module. Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention makes full use of visual technology, without touching or modifying any part of the fume hood, thus maintaining the integrity and safety of the equipment; In addition, the present invention uses directional bounding box detection, which overcomes the error caused by the tilt of the viewing angle and improves the positioning accuracy by more than 30%. By using delayed alarms and multi-level early warning mechanisms, the system effectively distinguishes between momentary unintentional behavior and continuous violations, thereby reducing human negligence and improving user experience. Attached Figure Description

[0016] Figure 1 This is a block diagram showing the overall architecture of the intelligent monitoring and alarm system for the height of a fume hood window based on visual target detection, as described in this invention. Figure 2 This is a flowchart illustrating the steps of the intelligent monitoring and alarm method for the height of the fume hood window based on visual target detection according to the present invention. Detailed Implementation

[0017] To provide a further understanding of the purpose, structure, features, and functions of the present invention, detailed descriptions are provided below with reference to specific embodiments.

[0018] Example 1: System Deployment and Core Algorithm Implementation refer to Figure 1 and Figure 2 This embodiment details the hardware connection and software algorithm process of the system, and focuses on explaining how each technical step achieves the beneficial effects of the present invention.

[0019] S1: System hardware initialization and calibration – achieving “easy deployment” and “flexible configuration”; The industrial camera is fixedly mounted in front of the fume hood and connected to the edge computing unit (NVIDIA Jetson NX) via a network cable. After the system is powered on, the calibration procedure is run: the operator accesses the calibration interface via the touchscreen, manually lowers the fume hood window to a physically safe height (e.g., 20 cm from the tabletop), and then, in the live view displayed on the interface, clicks sequentially on the left endpoint A and the right endpoint B of the bottom crossbeam of the window. The system records the pixel coordinates A(xa, ya) and B(xb, yb) of these two points, and the line connecting them forms a virtual safety line.

[0020] This calibration process is entirely completed at the software level, eliminating the need for installing any physical sensors or drilling holes in the fume hood rails. This solves the pain point of "modifying the fume hood structure" and enables convenient "plug-and-play" deployment. The virtual safety line is defined by software, rather than being fixed by hardware. For different experiment types, personnel heights, or fume hood models, administrators only need to recalibrate to adjust the safety height without any hardware modifications, greatly improving the system's adaptability.

[0021] S2: Real-time image acquisition and preprocessing – ensuring "fast response speed" and "strong anti-interference capability"; The camera captures RGB images at a frame rate of 15fps and pushes them to the edge computing unit. The image processing module performs preprocessing on each frame: Size scaling: The image is scaled to a fixed size of 640*640 required by the model. Fixed-size scaling significantly reduces the amount of data that the subsequent model needs to process, directly reducing the processing time per frame, and is the basis for achieving "end-to-end processing latency of less than 200ms".

[0022] Color normalization: Normalizes pixel values ​​from [0, 255] to [0, 1]. Pixel normalization eliminates the differences in the original image caused by lighting variations, providing a standardized input for the model and enabling it to maintain stable recognition performance when faced with common laboratory lighting variations.

[0023] S3: Precise window positioning based on the OBB model – a core innovation that achieves "high detection accuracy"; The preprocessed image is fed into the deployed YOLOv8-OBB model for inference. The model outputs the directional bounding box information of the bottom beam of the viewport, including the pixel coordinates of the four vertices (x1, y1), (x2, y2), (x3, y3), and (x4, y4) and the confidence score S. The system only processes detection results with a confidence score S > 0.7.

[0024] Unlike traditional AABB (Axis-Aligned Frames), which contain a lot of background interference when the viewpoint is tilted, the OBB model can closely fit the tilted window outline. Through precise fitting, the obtained window position information is more accurate, thereby reducing the average positioning error by more than 30% and directly solving the viewpoint tilt problem caused by installation position limitations.

[0025] S4: Window position calculation and boundary violation judgment - to achieve "rigorous judgment logic"; After receiving the OBB vertex coordinates, the security monitoring module calculates the coordinates C(x) of the viewport geometric center point based on the coordinates of the four vertices. c ,y c );

[0026] The center point represents the overall position of the current window and serves as the basis for subsequent line crossing judgments.

[0027] Use the cross product method to determine the center point C(x) c ,y c Regarding the positional relationship with respect to line AB, let:

[0028] Cross product calculation:

[0029] The judgment rules are as follows: • If cross > 0: The center point is above the safety line → crosses the line; • If cross < 0: The center point is below the safety line → safe; If cross=0: The center point is exactly on the safety line → This is considered a critical state, and a warning is advised.

[0030] S5: Intelligent delayed alarm triggering – ensuring "high level of intelligence" and "low false alarms"; The system internally maintains a violation counter. When a violation is detected, the counter increments; when a safe state is detected, the counter resets to zero. A final alarm is triggered only when the duration corresponding to the accumulated number of frames exceeds a preset delay threshold T_alert (e.g., 15 seconds).

[0031] Calculation formula: Violation duration (seconds) = (violation_counter / video frame rate (fps)) When the continuous out-of-bounds time is greater than or equal to T_alert, the alarm module is activated, and an audible and visual alarm, a screen pop-up, and a voice broadcast are executed.

[0032] This mechanism effectively filters out brief, unintentional boundary violations (such as people briefly peering into the device or reaching out to adjust it). The delay mechanism ensures that alarms are triggered for truly overlooked and persistent safety hazards, thereby greatly reducing false alarm rates and improving user experience and system reliability.

[0033] Example 2: Application of Dynamic Security Lines – Demonstrating an Extension of "Flexible Configuration" In another embodiment, the safety line can be dynamically adjusted according to the experimental risk level, further demonstrating the "configuration flexibility" of the system.

[0034] In the system configuration, a safety height adjustment coefficient k_risk is preset for different risk levels.

[0035] Low risk: k_risk = 1.0 (using standard safety height) High risk: k_risk = 0.7 After the operator selects the experiment risk level, the system dynamically calculates the new safety line Y-coordinate according to the following formula: y = y a +(y b -y a )×k_risk, where y aThe Y-coordinate of the bottom in the image coordinate system when the window is completely closed. b The Y-coordinate of the original security line in the image coordinate system.

[0036] The present invention has been described in the above-described embodiments; however, these embodiments are merely examples for implementing the present invention. It must be noted that the disclosed embodiments do not limit the scope of the present invention. Conversely, any modifications and refinements made without departing from the spirit and scope of the present invention are within the scope of patent protection of the present invention.

Claims

1. A fume hood window height intelligent monitoring and alarm system based on visual target detection, characterized in that: include The image acquisition module is used to continuously acquire video streams containing the viewing area of ​​the fume hood; An image processing module, connected to the image acquisition module, is used to preprocess the video stream images and identify the window position based on a deep learning object detection model; A safety monitoring module, connected to the image processing module, is used to make boundary judgments based on the identified window position and the preset safety height line. An alarm module, connected to the safety monitoring module, is used to trigger a multi-mode alarm signal when the safety monitoring module determines that a continuous boundary crossing has occurred. The system calibration module is used to establish the mapping relationship between the physical world safety height and the safety line in the image coordinate system.

2. The intelligent monitoring and alarm system for the height of a fume hood window based on visual target detection as described in claim 1, characterized in that: The image processing module uses a deep learning object detection model based on oriented bounding boxes, which is used to detect the bottom beam or handle of the viewport and output the coordinates of its bounding box vertices with rotation angle.

3. The intelligent monitoring and alarm system for the height of a fume hood window based on visual target detection as described in claim 2, characterized in that: The boundary judgment logic of the security monitoring module is as follows: using the coordinates of the upper edge vertex of the directional bounding box, the cross product method is used to determine whether the vertex is located above the preset safety line.

4. The intelligent monitoring and alarm system for the height of a fume hood window based on visual target detection as described in claim 1, characterized in that: The security monitoring module also includes a delay judgment unit, which, after detecting an out-of-bounds state, determines it as a valid violation and triggers the alarm module only if the state continues to exceed a preset delay threshold.

5. The intelligent monitoring and alarm system for the height of a fume hood window based on visual target detection as described in claim 1, characterized in that: The alarm module supports multi-level alarms, including: Level 1 warning: A visual alert is triggered when the viewing window height approaches the safety line. Level 2 alarm triggers sound, light, voice, and remote notifications when the viewing window height continuously exceeds the safety line.

6. The intelligent monitoring and alarm system for the height of a fume hood window based on visual target detection as described in claim 1, characterized in that: It also includes a system self-diagnosis module, which monitors the working status and image quality of the image acquisition module and triggers a system fault alarm when an abnormality occurs.

7. A method for intelligent monitoring and alarm of fume hood window height based on visual target detection, characterized in that: Applied to the system as described in any one of claims 1-6, the method comprises the following steps: S1: The system calibration module completes the calibration from the physical safety height to the image safety line; S2: The image acquisition module acquires video images of the fume hood viewing area in real time. S3: The image processing module preprocesses the image and uses a deep learning model to identify the precise location of the window; S4: The security monitoring module compares the identified window position with the image security line to determine whether an out-of-bounds violation has occurred. If a continuous out-of-bounds violation occurs, an alarm signal will be triggered through the alarm module.

8. The intelligent monitoring and alarm system for the height of a fume hood window based on visual target detection as described in claim 7, characterized in that: The "identifying the precise position of the viewport" specifically involves: using the YOLOv8-OBB model to detect the bottom features of the viewport and outputting the vertex coordinates of its directional bounding box; the "determining whether an out-of-bounds error has occurred" specifically involves: calculating the positional relationship between the upper edge vertex of the directional bounding box and the safety line.

Images