Cage overloading detection method based on machine vision

Abstract

The invention discloses a cage overloading detection method and device based on machine vision. The method and device are mainly applied to personnel safety monitoring in mines, construction sites and other places. According to the method, the personnel entering the cage are monitored in real time through a single intrinsic safety type camera, the number of the personnel in the cage is accurately detected and counted through a deep learning algorithm, and the complexity of traditional multi-camera linkage and equipment signal dependence is avoided. The system carries out target detection and tracking on personnel and a cage door by training YoloV10 and DeepSort models, starts counting logic according to the opening and closing state of the cage door, reminds overcrowded personnel to evacuate through voice alarm when the number of people entering the cage door exceeds a preset upper limit, and uploads detection data and videos to an AI platform for management personnel to check and process. The method does not need other equipment signal access, has the advantages of simple and convenient installation, accurate detection, strong robustness and the like, and can effectively improve the safety management level of the cage.

Description

Technical Field

[0001] The invention relates to the technical field of machine vision detection, and in particular to a tank cage overcrowding detection method based on machine vision. Background Art

[0002] With the development of the medical industry, the packaging requirements for medical drugs such as Band-Aids are becoming increasingly stringent. As a common topical drug, Band-Aids are usually used for the care and protection of small wounds. Therefore, their packaging not only needs to be moisture-proof and antibacterial, but also needs to ensure that the Band-Aids are not contaminated during long-term storage. Traditional Band-Aid packaging mostly uses paper boxes or ordinary plastic packaging. This type of packaging is easily affected by moisture in a high-humidity environment, especially in humid climatic conditions, such as the rainy season in the south. Moisture may penetrate into the packaging, causing the Band-Aid to lose its original sterile properties, affecting the use effect and safety. In addition, the sealing and airtightness inside the packaging are insufficient, and it is impossible to effectively prevent the invasion of external moisture and microorganisms, further affecting the preservation quality of the Band-Aid.

[0003] The packaging materials used for Band-Aids in the prior art still have deficiencies in terms of moisture-proof, antibacterial and airtightness. In particular, during long-term storage, the humidity control inside the package is difficult to maintain stability. The current packaging technology is difficult to ensure that the Band-Aids remain dry and sterile in a high-humidity environment, and the existing sealing process often cannot provide sufficient airtightness, resulting in the humidity level inside the package exceeding the safe range. Therefore, it is urgent to develop a new packaging process that can effectively solve the moisture-proof and antibacterial problems of Band-Aids during long-term storage and ensure their quality and safety under various environmental conditions. Summary of the invention

[0004] The present invention relates to a cage overcrowding detection method and device based on machine vision, which is specifically used in densely populated places such as mines and construction sites. The method aims to monitor and count the people entering the cage in real time by using advanced deep learning algorithms, so as to ensure the safe operation of the cage and prevent equipment damage and personal injury accidents caused by overcrowding.

[0005] The current methods for detecting overcrowding in tank cages mainly fall into the following categories: detection methods based on photoelectric sensors, personnel positioning cards, and image processing. Among them, the detection cost based on photoelectric sensors is relatively high, and the detection based on personnel positioning cards is easily affected by factors such as the battery level and wearing condition of the positioning cards, and has a low accuracy rate. Although the existing methods for detecting overcrowding in tank cages based on image processing have certain advantages in terms of easy installation and non-contact measurement, they often require the linkage of multiple cameras and need to be combined with relevant signals from tank cage equipment for detection. This method is complex to install, and it is difficult to adjust the angles of multiple cameras. The angle offset of a camera or the interruption of equipment signals may lead to detection errors, and its robustness is poor.

[0006] Therefore, the present invention proposes a detection method for identifying the overcrowding of a double cage using a single camera, without the need to access signals from other devices, thereby improving detection accuracy and robustness. The present invention uses a target detection and tracking model based on deep learning to accurately count the number of people entering the cage and trigger a voice alarm when overcrowded, thereby ensuring the safe use of the cage.

[0007] 1. Technical issues

[0008] The technical problem to be solved by the present invention is how to use a single camera, through machine vision technology and deep learning models, to monitor and count the people entering the cage in real time without relying on multi-camera linkage and external cage equipment signals, detect whether there is an overcrowding situation, and notify on-site staff through an alarm system to take measures to ensure safe operation in the cage.

[0009] 2. Technical solution

[0010] In order to solve the above technical problems, the present invention provides a method and device for detecting overcrowding in a tank cage based on machine vision. The core technical solution of the method includes the following steps:

[0011] Image acquisition: An intrinsically safe camera installed at the cage wellhead is used to collect images of people entering the cage. The camera resolution is not less than 1280×720, and the camera is installed at a height of not less than 3.5 meters. It is also installed at a 45-degree angle to the wellhead to ensure that the camera can clearly capture the head and shoulder information of people entering and exiting, which is convenient for subsequent personnel detection and counting.

[0012] Training of target detection and tracking models: Using the collected image dataset of personnel in the cage area, the combined model of YoloV10+DeepSort is used for target detection and tracking. Specifically, YoloV10 achieves more efficient target detection by introducing a dual allocation strategy, large kernel convolution, and partial self-attention (PSA) modules; DeepSort enhances the robustness of personnel tracking by fusing features and trajectory prediction. After training, the system can accurately detect and track each person entering the cage, and assign a unique TrackID to each person to avoid duplicate counting.

[0013] Detection and judgment of cage door status: Set the ROI area in the image to monitor the opening and closing status of the cage door. Calculate the opening and closing status of the cage door by monitoring the change of cage door pixel position in real time. When the cage door is opened, start the counting logic of the personnel entering the well and start tracking the personnel entering the cage in real time.

[0014] Counting of personnel entering the cage: When the cage door is opened, the system uses the ray method to detect and count the personnel position according to whether the personnel's upper midpoint coordinates have entered the cage area. Each person entering the cage will be marked as having entered the well by the system to avoid repeated counting. At the same time, real-time counting is performed according to the preset upper limit of the cage's carrying capacity.

[0015] Overcrowding alarm: When the number of people entering the cage exceeds the preset upper limit, the system will trigger the alarm mechanism, play a voice reminder, notify the excess people to leave the cage, and at the same time save the surveillance video with personnel marks and upload it to the AI ​​platform for management personnel to view.

[0016] Data storage and processing: The system can store real-time collected images and personnel tracking data. The data storage device includes an SD card or hard disk with a large storage capacity, so that subsequent management personnel can replay and analyze the video and data.

[0017] 3. Beneficial effects

[0018] The present invention uses machine vision and deep learning technology to complete the detection and alarm of cage overcrowding using a single camera, which has the following beneficial effects:

[0019] The equipment is easy to install: no complicated multi-camera linkage and angle adjustment are required, which reduces the complexity of equipment installation and maintenance.

[0020] Accurate detection: The deep learning-based target detection and tracking model can achieve accurate identification and counting of people, avoiding the inaccurate detection problems caused by equipment failure or people's negligence in wearing positioning cards in traditional sensor methods.

[0021] High robustness: It can independently complete the detection task without relying on the relevant signals of the cage equipment and is not affected by the interruption of the equipment signal.

[0022] Real-time alarm: When the tank cage is overcrowded, the system can immediately remind on-site personnel through the voice alarm device to ensure timely handling of the overcrowding situation and ensure personnel safety.

[0023] Data storage and backtracking: The system can save overcrowding detection data and videos for subsequent dispatch personnel to analyze and trace, facilitating management.

[0024] The tank cage overcrowding detection method and device based on machine vision proposed in the present invention not only improve the detection accuracy and robustness of the system, but also simplify the installation and maintenance process, and have high practicality and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 A schematic diagram of a workflow diagram of an embodiment of the present invention;

[0027] Figure 2 Schematic diagram of camera installation angle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0029] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0030] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0031] Example 1

[0032] The present invention proposes a method and device for detecting overcrowding of personnel entering a cage based on machine vision. As shown in the figure below, the invention uses an intrinsically safe camera to pre-collect images of personnel entering the cage and establish an image data set. The target detection and tracking model is trained using the images in the image data set to monitor and track two types of targets: personnel and cage doors. The current state of the cage is monitored in real time, and the intersection and ratio calculation is performed using the pixels of the cage door detected in real time and the marked cage area. The opening and closing state of the cage is judged according to the set threshold and the position of the cage opening is determined (left and right cages). Taking the left cage as an example, when the cage is opened, the video saving logic is started, the personnel entering are tracked, and the number of people entering this time is counted. When the number of people entering the cage exceeds the limit, the on-site linkage camera sends a voice alarm to remind the overcrowded personnel entering the well to exit the cage and return to the waiting area. At the same time, the marked well entry video is saved and uploaded to the AI ​​platform, which is convenient for the dispatching operation and maintenance personnel to obtain the alarm information in time and deal with the on-site problems.

[0033] The present invention uses an intrinsically safe camera to collect images, and the resolution is not less than 1280*720, and the camera is installed at a height greater than 3.5 meters. Figure 2 As shown, the camera is at a 45-degree angle to the wellhead and is located on the opposite side of the screen from where the personnel enter and exit, so that the personnel's entry and exit routes can be clearly seen. The wellhead area occupies about 1 / 2 of the entire screen, and the camera screen can fully see the personnel's head and shoulders, with no obstruction to the front, back, left, and right heads and shoulders of the personnel; if personnel can enter and exit the well from both sides, the camera should cover the areas on both sides of the wellhead.

[0034] (2) Training target tracking model

[0035] The collected images are annotated, and two categories, personnel and cage doors, are annotated to train the target detection model. Since the processing of the collected images in the present invention is completed at the edge, YoloV10+DeepSort is used as the detector and tracker. YoloV10 introduces a dual allocation strategy, eliminating the need for NMS, thereby achieving faster and more efficient target detection. And by integrating large kernel convolution and partial self-attention (PSA) modules, the model detection accuracy is improved. Therefore, YoloV10 can run with real-time speed and high-precision reasoning on small edge devices. As a classic target tracker, DeepSort realizes the fusion analysis of target features and trajectory prediction, and improves the robustness of tracking. Personnel and cage doors are detected by the target detection model, and the personnel detection category is sent to the target tracking model to track the personnel entering the cage in detail, and unique TrackIDs are assigned to different personnel. The cage door detection category is used to subsequently determine whether the cage is open.

[0036] It should be noted that no matter which neural network model is used to achieve target tracking of the acquired image, it does not affect the essence of this solution.

[0037] (3) Determine the cage opening and closing status

[0038] Set the ROI area so that the image ROI area contains the area where the cage is completely closed. The upper left corner and lower right corner of the ROI have position coordinates of [X1, Y1, X2, Y2]. Detect the current cage door position information in the image. The position information is [x1, y1, x2, y2]. Set a reasonable threshold a. If |X1-x1|>a, the left cage is open. If |X2-x2|>a, the right cage is open.

[0039] (4) Counting of people entering the cage

[0040] When it is detected that the cage is open, the tank entry counting logic is activated. Taking the right cage as an example, when the right cage is opened, the well entry judgment area corresponding to the right cage takes effect. If the coordinates of the upper midpoint of the detected person are generated in the waiting area, the tracked person id is marked as the object to be entered into the well. The present invention uses the ray method to determine whether the point is within the area. If the id is tracked through the cage entry area, the well entry person +1. At this time, the tracking id is marked as having entered the well, and the id will not be counted repeatedly when it wanders between the two areas later. The coordinates of the upper center point of the detected person are generated in the cage entry area. Only the person id is tracked and not counted.

[0041] (5) Online detection process

[0042] By collecting images offline, establishing an image data set, and training an image target detection model. The training of this model is completed offline, so sufficient learning and training can be carried out. When implementing online detection, the image of the cage area is collected through an intrinsically safe camera, and the image is input into the target detection model and target tracking model in step 2 to obtain the cage door detection and personnel tracking results of the image. Then, on this basis, the number of people entering the cage is counted according to steps 3 and 4. The upper limit of personnel requirements for the tank entry areas on both sides of the cage are different, and the system can set the upper limit of the number of people respectively. Before the cage is closed, if the number of people entering this time exceeds the set number threshold, an alarm will be issued, and the camera will be linked to make a voice broadcast to remind the overcrowded personnel on site to leave the cage.

[0043] (6) The equipment

[0044] A tank cage overcrowding detection method and device based on machine vision, including image acquisition equipment, communication transmission equipment, data processing equipment, data storage equipment, etc.

[0045] The image acquisition equipment is an intrinsically safe camera in coal mines, which can collect high-definition images of the transmission belt during operation. The surface of the camera needs to be made of dust-proof material to avoid the adhesion of dust in the mine, which will affect the image quality.

[0046] Communication transmission equipment includes equipment required for wireless transmission methods such as 5G and industrial wifi, as well as equipment required for wired transmission methods such as USB3.0 and Ethernet.

[0047] The edge processing device uses an AI edge computing server, which has a certain amount of computing power and memory to complete image target detection based on the YoloV10 model and target tracking tasks based on DeepSort.

[0048] The data storage device is an SD card or a hard disk with a large storage capacity, which is used to store the images identified as deviations.

[0049] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0050] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A cage overcrowding detection method based on machine vision, characterized in that: The following steps are involved: Step 1: Image acquisition: Use intrinsically safe cameras to collect images of people entering the cage. The camera resolution should be no less than 1280×720, and the installation height should be greater than 3.5 meters. The camera should be at a 45-degree angle to the wellhead entrance and exit direction to ensure that the camera can fully cover the cage entrance and exit area and capture the head and shoulders of people entering the cage without obstacles. Step 2: Model training: By annotating the pre-collected images of people entering and exiting the cage, an image dataset including two types of targets, people and cage doors, is established. YoloV10+DeepSort is used as the target detection and tracking model, and the model is trained to detect the cage door status and track and count the people entering. Among them, YoloV10 introduced a dual allocation strategy to improve detection accuracy by integrating large kernel convolution and partial self-attention (PSA) modules, while DeepSort enhanced the robustness of personnel tracking by integrating target features and trajectory prediction. Step 3: Determine the cage door status: Set ROI (region of interest) to detect the cage door position information in the image, and determine the open or closed state of the cage door by calculating the intersection-over-union ratio of the image. When the cage door is determined to be open, start the video monitoring and personnel tracking functions; Step 4: People counting: When the cage door is opened, the entry counting logic is started to track the personnel entering the cage area and count the number of personnel entering the cage. Each tracked person is given a unique TrackID to avoid duplicate counting. When the detected personnel enter the cage and the number of people exceeds the preset upper limit of the cage, the system prompts the excess personnel to leave the cage through the voice alarm device and saves the marked video data for subsequent query; Step 5: Alarm linkage: When it is detected that the number of people in the cage exceeds the set maximum value, the alarm system is activated, a voice broadcast reminder is given to prompt the excess people to evacuate, and the alarm video is uploaded to the AI ​​platform for the dispatcher to view; Step 6: Data storage: The relevant data and images of overcrowding detection in the cage are stored in the SD card or hard disk of the local device for tracing back and analysis when needed.

2. The method for detecting overcrowding in a cage based on machine vision according to claim 1, characterized in that: The surface of the intrinsically safe camera is made of dust-proof material to prevent the dust in the mine from affecting the imaging quality.

3. The method for detecting overcrowding in a cage based on machine vision according to claim 1, characterized in that: The communication transmission equipment includes wireless transmission equipment and wired transmission equipment. The wireless transmission equipment includes 5G and industrial WiFi, and the wired transmission equipment includes USB3.0 and Ethernet.

4. The method for detecting overcrowding in a cage based on machine vision according to claim 1, characterized in that: The upper threshold of the number of people in each cage in the system can be independently set according to the carrying capacity of different cages.

5. The method for detecting overcrowding in a cage based on machine vision according to claim 1, characterized in that: The detection model can be inferred in real time on-site through an AI edge computing server, reducing data transmission delays.

6. A tank cage overcrowding detection device based on machine vision, characterized in that: It includes image acquisition equipment, communication transmission equipment, data processing equipment and storage equipment, among which: The image acquisition device is an intrinsically safe camera; Communication transmission equipment includes wireless transmission devices such as 5G and industrial WiFi, and wired transmission devices such as USB3.0 and Ethernet; The data processing device is an AI edge computing server, which is responsible for image target detection based on the YoloV10 model and target tracking based on DeepSort; The data storage device includes an SD card or a hard disk for storing relevant images and data.

Images