People clamping accident detection method and system, storage medium and computer equipment

Abstract

The invention discloses a person clamping accident detection method and system, a storage medium and computer equipment. The invention relates to a person clamping accident detection method, which comprises the following steps: acquiring an RGB image and a thermal imaging image when an automatic curtain door is closed, and detecting a human body target at the closing position of the automatic curtain door in the RGB image; in response to a human body target detected at the closing position of the automatic curtain door, marking the human body target in the thermal imaging image, and identifying each key skeleton point of the human body target to form a skeleton frame diagram of the human body target; carrying out temperature sampling on the human body target in the thermal imaging image based on the skeleton frame diagram to obtain the temperature of the human body target; and in response to the fact that the temperature of the human body target is within the preset body temperature threshold range, reminding a person clamping accident. In this way, the misjudgment rate that people are clamped by the automatic curtain door can be reduced.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method, system, storage medium and computer equipment for detecting human-entrapment accidents. Background Art

[0002] In traditional detection methods, the technology of human-entrapment detection mainly relies on traditional visual monitoring systems, especially RGB cameras, which perform feature recognition through RGB images to detect whether there is a human body at the door closing position, thereby determining the occurrence of a human-entrapment accident. However, in complex environments, especially in the application scenarios of transparent curtain doors, it is easily disturbed by light changes and reflections. For example, when a passenger is waiting in front of a transparent curtain door, his or her reflection may be misjudged as a person trapped in the door gap, resulting in a false alarm and a high false detection rate of human-entrapment in automatic doors. Summary of the invention

[0003] The present application mainly provides a method, system, storage medium and computer equipment for detecting people-pinching accidents, so as to solve the problem of high false detection rate of people-pinching accidents caused by automatic doors.

[0004] In order to solve the above technical problems, a technical solution adopted in the present application is: to provide a method for detecting a person-pinching accident, comprising: obtaining an RGB image and a thermal imaging image when an automatic curtain door is closed, and detecting a human target at the closing position of the automatic curtain door in the RGB image; in response to detecting a human target at the closing position of the automatic curtain door, marking the human target in the thermal imaging image, and identifying each joint point of the human target to form a skeletal framework diagram of the human target; performing temperature sampling on the human target in the thermal imaging image based on the skeletal framework diagram to obtain the temperature of the human target; in response to the temperature of the human target being within a preset body temperature threshold range, an alert is given that a person-pinching accident has occurred.

[0005] In some embodiments, marking the human target in the thermal imaging image and identifying the joints of the human target to form a skeletal framework diagram of the human target includes: marking the human target in the thermal imaging image after alignment with the RGB image through a target detection model; identifying the joints of the human target through a key point detection algorithm to form a skeletal framework diagram of the human target.

[0006] In some embodiments, temperature sampling of the human target in the thermal imaging image based on the skeletal framework diagram to obtain the temperature of the human target includes: temperature sampling of each key bone point in the skeletal framework diagram; and calculating the temperature of the human target based on the temperature sampling results of each key bone point in the skeletal framework diagram.

[0007] In some embodiments, the temperature sampling result of the human target is calculated based on the temperature of each key bone point of the skeletal framework diagram, including: calculating the weighted average temperature of the human target based on the temperature sampling results of each key bone point of the skeletal framework diagram and the corresponding temperature weights.

[0008] In some embodiments, obtaining the RGB image and the thermal imaging image when the automatic curtain door is closed includes: collecting the RGB image and the thermal imaging image at the same viewing angle when the automatic curtain door is closed through a binocular camera; and aligning the RGB image and the thermal imaging image by establishing a mapping relationship between corresponding feature points in the RGB image and the thermal imaging image.

[0009] In some embodiments, before detecting the human target at the closed part of the automatic curtain door in the RGB image, it also includes: acquiring RGB images of the automatic curtain door in different switch states and different lighting environments, and annotating the acquired RGB images to construct a training set and a verification set; training a deep learning-based target detection algorithm model through the training set; verifying the performance of the target detection algorithm model with the verification set; and using the RGB image containing the automatic curtain door as input of the trained target detection algorithm model to detect the switch state of the automatic curtain door.

[0010] In some embodiments, the training of the deep learning-based target detection algorithm model using the training set includes: optimizing the parameters of the target detection algorithm model through back propagation and gradient descent algorithms; and ending the training in response to the convergence of the loss function constructed by the target detection algorithm model during the back propagation process.

[0011] In order to solve the above technical problems, another technical solution adopted by the present application is: providing a storage medium on which program data is stored, and when the program data is executed by a processor, the steps of the above-mentioned method for detecting a person-pinching accident are implemented.

[0012] The present application also provides a computer device, including a processor and a memory connected to each other, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned method for detecting a person-pinching accident are implemented.

[0013] The present application also provides a system for detecting a person-pinching accident, comprising a binocular camera and a computer device as described above; the binocular camera is installed above the automatic curtain door, and is used to continuously acquire RGB images and thermal imaging images and send them to the computer device; the computer device receives the RGB images and thermal imaging images sent by the binocular camera to execute the method for detecting a person-pinching accident.

[0014] The beneficial effects of the present application are as follows: Different from the prior art, the present application discloses a method, system, storage medium and computer device for detecting a person-pinching accident. The RGB image and thermal imaging image of the automatic curtain door when it is closed are obtained, and the human target at the closed part of the automatic curtain door in the RGB image is detected to identify possible person-pinching accidents; in response to the detection of the human target at the closed part of the automatic curtain door, the human target is marked in the thermal imaging image, and the joints of the human target are identified to form a skeleton frame diagram of the human target, and the temperature of the human target in the thermal imaging image is sampled based on the skeleton frame diagram to obtain the temperature of the human target; in response to the temperature of the human target being within the preset body temperature threshold range, a person-pinching accident is reminded to avoid the influence of the glass reflection or the human shadow behind the door on the automatic door person-pinching detection, and to avoid the misjudgment of the person-pinching accident. The combination of visual information and thermal imaging technology not only improves the recognition accuracy of the person-pinching accident, but also enhances the ability to handle potential false alarms in complex environments; through real-time analysis, an alarm can be issued in time to effectively ensure the safety of passengers, providing strong technical support for the safety management of the public transportation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which: Figure 1 It is a flow chart of an embodiment of a method for detecting a person-entrapment accident provided by the present application; Figure 2 Yes Figure 1 The flowchart of the embodiment of step 10 of the method shown is as follows; Figure 3 Yes Figure 1 The flowchart of the embodiment of step 20 of the method is shown; Figure 4 Yes Figure 3 The flowchart of the embodiment of step 20 of the method is shown; Figure 5 Yes Figure 4 The camera image of step 32 of the illustrated method in an embodiment; Figure 6 It is a flow chart of another embodiment of the method for detecting a person-entrapment accident provided by the present application; Figure 7 Yes Figure 5 The flowchart of the embodiment of step 02 of the method shown; Figure 8It is a structural schematic diagram of an embodiment of a storage medium provided by the present application; Fig. 9 is a structural schematic diagram of an embodiment of a computer device provided by the present application; Fig.10 It is a structural schematic diagram of an embodiment of a human trap accident detection system provided by the present application. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0017] The terms "first", "second", "third" in the embodiments of the present application are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first", "second", "third" can expressly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0018] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0019] See also Figure 1 , Figure 1 1 is a flow chart of an embodiment of a method for detecting a person-pinching accident provided by the present application. The method for detecting a person-pinching accident includes: Step 10: Obtain the RGB image and thermal imaging image when the automatic curtain door is closed, and detect the human target at the closing position of the automatic curtain door in the RGB image.

[0020] Capture color images and thermal imaging images when the curtain door is closed, detect human targets in the RGB image by processing and analyzing image data, analyze the shape and position of the detected human targets, and determine whether there is a suspected human trapping accident when the curtain door is closed.

[0021] Human targets are targets with human features obtained through image feature recognition based on RGB images. They may be real human bodies, human reflections, or perspective human figures.

[0022] Further, see Figure 2 , step 10 comprises: Step 11: Use the binocular camera to collect RGB images and thermal imaging images from the same viewing angle when the automatic curtain door is closed.

[0023] Capture RGB and thermal images simultaneously with a binocular camera. The binocular camera contains two side-by-side sensors: a standard RGB sensor for capturing color images and a thermal sensor for capturing thermal radiation images of objects. Install the binocular camera system in an appropriate position to capture the complete door area and possible pedestrians when the automatic door is closed. Adjust the camera's exposure, focal length, frame rate and other parameters as needed to ensure that the image quality meets the analysis requirements. Start the camera system and begin collecting RGB and thermal images when the automatic door is closed. Ensure that the image data can be saved to the specified storage device in real time or on demand.

[0024] Step 12: Align the RGB image and the thermal image by establishing a mapping relationship between corresponding feature points in the RGB image and the thermal image.

[0025] Since RGB images and thermal images are captured based on different physical principles, there may be slight perspective differences or image distortions between them. In order to accurately align the two images, it is necessary to establish a mapping relationship between the corresponding feature points between them.

[0026] Specifically, feature points are detected in the RGB image and the thermal image respectively. These feature points can be corner points, edge points or areas with significant texture in the image. Use feature matching algorithms (such as SIFT, SURF, ORB, etc.) to find corresponding feature points between the RGB image and the thermal image. These algorithms find the best matching point pairs by comparing the descriptors of the feature points. Based on the matched feature point pairs, the transformation matrix from the RGB image to the thermal image is calculated. This matrix describes how to map the points in the RGB image to the corresponding points in the thermal image. Apply the transformation matrix to transform the RGB image so that it is aligned with the thermal image, including translation, rotation, scaling and / or affine transformation of the image. Check the aligned image to ensure that the corresponding feature points in the RGB image and the thermal image have been accurately aligned.

[0027] This sophisticated alignment process helps to more effectively extract and fuse the information of the two images in subsequent analysis, thereby providing more accurate and reliable data support for human pinching detection.

[0028] Step 20: In response to detecting a human target at the closing position of the automatic curtain door, the human target is marked in the thermal imaging image, and each joint point of the human target is identified to form a skeleton framework diagram of the human target.

[0029] Thermal imaging can effectively distinguish real human bodies from other objects by detecting temperature changes in the target area. The system will perform temperature analysis on the detected target and combine the thermal characteristics of different parts of the human body to ensure the accuracy of the judgment. First, the target detection algorithm is used to identify the human body in the thermal imaging image, and a corresponding bounding box is generated for each detected human body. Subsequently, the system further analyzes the posture of the human body and identifies the key bone points of the human body, such as the head, shoulders, elbows, wrists, hips, knees, and ankles, by introducing a key point detection algorithm. By detecting these joint points, a skeleton framework diagram of the human body is formed, which provides a basis for subsequent temperature analysis.

[0030] Further, see Figure 3 , step 20 comprises: Step 21: Use the target detection model to mark human targets in the thermal imaging image after alignment with the RGB image.

[0031] Human targets are identified in thermal images using deep learning-based human detection algorithms such as the YOLO algorithm. A bounding box is generated for each detected human target to mark its location in the image.

[0032] Step 22: Identify the joints of the human target through a key point detection algorithm to form a skeleton framework diagram of the human target.

[0033] The deep learning-based human posture recognition algorithm is introduced to estimate the posture of the detected human target. The algorithm will identify the key bone points of the human body, such as the head, neck, shoulders, elbows, wrists, hips, knees and ankles. These key bone points will be marked on the thermal imaging image. According to the detected joint points, the algorithm will automatically connect the adjacent key bone points to form a skeleton diagram of the human body.

[0034] The detected skeleton frame diagram is superimposed on the thermal imaging image in real time so that the screen door control system can intuitively see the posture information of the human body.

[0035] By detecting these joints, a skeleton diagram of the human body is formed, which provides a basis for subsequent temperature analysis. This combination of visual information and thermal imaging technology not only improves the system's recognition accuracy of human-entrapment incidents, but also enhances its ability to handle potential false alarms in complex environments.

[0036] Step 30: Perform temperature sampling on the human target in the thermal imaging image based on the skeleton frame diagram to obtain the temperature of the human target.

[0037] The pixel values ​​in the thermal imaging image are converted into temperature values ​​based on the skeleton frame diagram, and the temperature value of the corresponding pixel is extracted for each key skeleton point located. This may require mapping the key skeleton point position to the pixel coordinates of the thermal imaging image and reading the temperature value at the coordinates. The extracted temperature data is statistically analyzed to determine whether the human target is a real human body based on the preset temperature threshold or the normal temperature range of the human body.

[0038] Further, see Figure 4 , step 30 comprises: Step 31: Perform temperature sampling on each key bone point in the skeleton frame diagram.

[0039] First, call the human skeleton model such as yolo-pose to generate the key skeleton points of the human target in the skeleton frame diagram. Convert the coordinates of each key skeleton point in the skeleton frame diagram from the output format of the pose estimation to the pixel coordinates of the thermal imaging image. Then, use the mapping relationship from the pixel value of the thermal imaging image to the temperature to extract the temperature value of the pixel position corresponding to each key skeleton point. Finally, record the temperature value of each key skeleton point.

[0040] Step 32: Calculate the temperature of the human target based on the temperature sampling results of each key bone point in the skeleton frame diagram.

[0041] Calculate the arithmetic mean of the temperature sampling results of all key bone points, that is, the temperature values, as the overall temperature of the human target.

[0042] Optionally, the weighted average temperature of the human target is calculated based on the temperature sampling results of each key bone point in the skeleton framework diagram and the corresponding temperature weight.

[0043] The weighted average of the temperature values ​​of all key bone points is calculated as the overall temperature of the human target. According to the importance or thermal contribution of different parts of the human body, a weight is assigned to the temperature value of each key bone point, and then the weighted average is calculated as the overall temperature of the human target.

[0044] Since the temperature of different parts of the human body varies, a corresponding weight is assigned to each key bone point. These weights are assigned based on the temperature characteristics of different parts under normal physiological conditions. For example, the temperature of the neck and armpits is usually higher, while the temperature of parts such as wrists and ankles is relatively lower. The introduction of this weight mechanism allows the importance of each part to be taken into account when calculating the overall temperature of each detected human body, thereby improving accuracy.

[0045] For example, see Figure 5 , Figure 5 The images in the figure are the platform curtain door and subway door taken by the RGB camera and the thermal imaging image taken by the thermal imaging camera at the same location.

[0046] Based on the target recognition algorithm model, the platform screen door is detected to be closed and the subway door is open. The subway door and platform screen door are set to be closed at the same time as the automatic screen door closed state, and the detection box marking the open and closed state of the platform screen door and subway door is displayed in the figure.

[0047] Based on the target detection model, the three human targets appearing in the picture are identified, which are located behind the subway door, between the two doors of the subway door, and at the glass of the platform door, and the position boxes of the three human targets are marked in the picture.

[0048] At this time, the subway door in the picture is in an open state, the platform curtain door is in a closed state, and a human reflection appears on the door. The reflection is identified as a human target and displayed in the RGB image and the thermal imaging image.

[0049] After generating the corresponding bounding box for each detected human body, the posture of the human body is further analyzed. By introducing the key point detection algorithm, the key bone points of the human body, such as the head, shoulders, elbows, wrists, hips, knees and ankles, are identified. By detecting these key bone points, a skeleton diagram of the human body is formed and displayed at the corresponding position of each human target, as shown in the RGB image on the left.

[0050] After obtaining the human skeleton framework, the system will use the body temperature characteristics of different parts of the human body to make a more accurate temperature estimate, and display it in different colors at the corresponding positions of the skeleton framework diagram. As shown in the thermal imaging image on the right, key bone points with different temperatures are displayed as lines or nodes of different depths to reflect the temperature of various target parts of the human body.

[0051] This process not only improves the accuracy of temperature judgment, but also effectively distinguishes between real pinching incidents and misjudgments caused by reflections or other passengers in the car. Thermal imaging technology plays a vital role in this, providing more reliable information through temperature differences.

[0052] Step 40: In response to the temperature of the human target being within a preset temperature threshold range, a reminder of a human-pinch accident is given.

[0053] The presence of a transparent door may cause refraction and reflection of light, and the system may mistakenly identify other objects or images as human bodies. For example, when other passengers approach the door, the transparent door may produce an image similar to a human body, causing a misjudgment. In addition, strong ambient light conditions, such as direct sunlight or strong artificial lighting, can also affect the accuracy of the RGB image, causing the system to mistakenly identify objects in the background as human bodies.

[0054] Thermal imaging can effectively distinguish between real human bodies and other objects by detecting temperature changes in the target area. The system will perform temperature analysis on the detected target and combine the thermal characteristics of different parts of the human body to ensure the accuracy of the judgment.

[0055] When the screen door is in the process of closing or about to close, and the thermal imaging camera detects that a human target is in the potential pinching area of ​​the screen door, the operator or nearby personnel are warned by sound, light or other visual / auditory prompts.

[0056] Optionally, see Figure 6 , before step 10, further comprising: Step 01: Obtain RGB images of the automatic curtain door in different switch states and different lighting environments, and annotate the obtained RGB images to construct training sets and validation sets.

[0057] Collect RGB images of automatic curtain doors in different switch states (such as open, closed, half-open, etc.) and different lighting environments (such as day, night, indoor lighting, outdoor natural light, etc.), and accurately annotate these images for subsequent use in training target detection algorithm models.

[0058] Use high-resolution cameras to capture the various states of automatic curtain doors at different time periods (e.g. morning, noon, evening, night) and under different lighting conditions (natural light, indoor light, mixed light, etc.) and ensure that the images cover all possible positions of the curtain doors (e.g. fully open, fully closed, and any position in between).

[0059] The key areas in each image are annotated, which may include the screen door itself, the edge of the screen door, and possible human targets.

[0060] The labeled image dataset is divided into a training set and a validation set. The training set is used to train the target detection algorithm model, and the validation set is used to evaluate the performance of the model.

[0061] Optionally, the training set occupies a larger proportion of the dataset (e.g., 70%-80%), while the validation set occupies a smaller proportion (e.g., 20%-30%).

[0062] The switch status of the automatic curtain door is detected through the target detection algorithm model to prevent accidents of people being trapped.

[0063] Step 02: Train the deep learning-based object detection algorithm model using the training set.

[0064] Use the training set constructed in step 01 to train the target detection algorithm model so that it can accurately identify and locate the automatic curtain door and its status.

[0065] Configure the model's parameters, such as input image size, batch size, learning rate, number of iterations, etc., and adjust the model's hyperparameters according to the characteristics of the dataset to optimize performance.

[0066] Input the training set into the algorithm model to start the training process. Monitor indicators such as loss function value and accuracy during the training process to evaluate the performance of the model. Adjust the parameters and configuration of the model as needed to improve performance.

[0067] Furthermore, if Figure 7 As shown, step 02 includes: Step 021: Optimize the parameters of the target detection algorithm model through back propagation and gradient descent algorithms.

[0068] The input data is forward propagated through the model to calculate the model's output. The loss function value is calculated based on the model's output and the true label. This value measures the quality of the model's prediction results. Then, the gradient of the loss function is back-propagated from the output layer to the input layer through the chain rule, that is, the partial derivative of the loss function with respect to each layer parameter is calculated. The gradient obtained by back-propagation is used to update the model parameters according to the specified learning rate.

[0069] Step 022: In response to the convergence of the loss function constructed by the target detection algorithm model during the back-propagation process, the training is terminated.

[0070] During the training process, as the model parameters are updated, the loss function value will gradually decrease. When the change in the loss function value is very small and close to a preset threshold, it can be considered that the loss function has converged. Once the loss function converges, it means that the performance of the model on the current data set has been relatively stable, and further training may not bring significant performance improvements, but may lead to overfitting. Therefore, you can stop training at this time and save the model parameters.

[0071] Through back propagation and gradient descent algorithms, the target detection algorithm model can continuously optimize its parameters to minimize the loss function. When the loss function converges, it means that the model has achieved good performance on the current data set, and the training can be terminated at this time.

[0072] Step 03: Verify the performance of the target detection algorithm model using the validation set.

[0073] The images in the validation set are input into the trained object detection algorithm model. The model processes each image and outputs the detected object (such as an automatic curtain door) and its location information.

[0074] Use standard evaluation metrics (such as precision, recall, F1 score, intersection over union, etc.) to measure the performance of the model on the validation set.

[0075] Analyze the performance of the model on the validation set to identify possible errors and deficiencies. If the model performance is poor, you may need to adjust the model structure, add data enhancement strategies, adjust hyperparameters, etc.

[0076] By using an independent validation set to evaluate the performance of the model, its accuracy and robustness in practical applications are ensured, thereby providing reliable data support for pinch point detection and improving passenger safety.

[0077] Step 04: Use the RGB image containing the automatic curtain door as the input of the trained object detection algorithm model to detect the open and closed status of the automatic curtain door.

[0078] Capture RGB images of automatic doors, ensure that the images are clear and contain sufficient context information. Preprocess the input images, such as resizing and normalizing, to match the input format for model training.

[0079] The preprocessed image is input into the trained target detection algorithm model, which processes the image and outputs the detected automatic curtain door and its location information.

[0080] According to the position information output by the model, the position of the automatic screen door in the image is determined, and the switch state of the screen door (such as open, closed, half-open, etc.) is judged by analyzing the position, size or shape of the screen door.

[0081] See also Figure 8 , Figure 8 It is a structural diagram of an embodiment of the storage medium provided by the present application.

[0082] The storage medium 300 stores program data 310. When the program data 310 is executed by the processor, the following is achieved: Figure 1 The steps of the method for detecting a person-entrapment accident are described.

[0083] The program data 310 is stored in a storage medium 300, and includes a number of instructions for enabling a network device (such as a router, a personal computer, a server, etc.) or a processor to execute all or part of the steps of the methods of the various embodiments of the present application.

[0084] Optionally, the storage medium 300 may be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or other medium that can store program data.

[0085] See also Fig. 9 , Fig. 9 It is a structural schematic diagram of an embodiment of a computer device provided by the present application.

[0086] The computer device 400 includes a processor 420 and a memory 410 which are connected to each other. The memory 410 stores a computer program. When the processor 420 executes the computer program, the steps of the above-mentioned method for detecting a person-entrapment accident are implemented.

[0087] See also Fig.10 , Fig.10 1 is a schematic diagram of the structure of an embodiment of a system for detecting a person-pinching accident provided in the present application. The system for detecting a person-pinching accident includes a binocular camera 500 and the computer device 400 as described above.

[0088] The binocular camera 500 is installed above the automatic curtain door and is used to continuously acquire RGB images and thermal imaging images and send them to the computer device 400 .

[0089] This location above the automatic door is chosen to ensure that the camera can capture a full view of the door area, thereby effectively monitoring any possible pinching situations. The camera has binocular vision capabilities, which means that it can simultaneously acquire RGB images and thermal images. RGB images provide visible light information in the door area, which helps to identify objects and people. Thermal images capture infrared radiation emitted by objects, which is particularly effective in detecting human heat and can accurately identify people even in low-light or completely dark environments. The binocular camera 500 sends the continuously collected image data in real time to the computer device 400 for further processing and analysis.

[0090] The computer device 400 receives the RGB image and the thermal imaging image sent by the binocular camera 500 to execute the above-mentioned method for detecting a person-entrapment accident.

[0091] The computer device 400 is responsible for receiving the RGB images and thermal imaging images from the binocular camera 500. These image data are the basis for subsequent detection of human entrapment accidents. The computer device 400 has a built-in advanced method for detecting human entrapment accidents. This method combines image processing, artificial intelligence and machine learning technologies, and is able to analyze the received image data and identify potential human entrapment situations. Specifically, the method may involve object recognition and tracking of RGB images, while using temperature information in thermal imaging images to enhance the accuracy of human detection. Once the computer device 400 detects a potential human entrapment accident, it will immediately trigger an early warning mechanism. This may include sending an alarm signal to the operator, automatically stopping the operation of the screen door, or taking other safety measures to prevent the occurrence of accidents.

[0092] The trapping accident detection system realizes effective monitoring and early warning of the automatic door area through continuous image acquisition of the binocular camera 500 and advanced image processing technology of the computer device 400. This system not only improves safety, but also reduces potential risks and losses caused by trapping accidents.

[0093] Different from the existing technology, this application combines RGB images with thermal imaging images, utilizes advanced multimodal data processing, and uses thermal imaging temperature detection to eliminate reflections and human figures behind doors that may be misjudged in RGB images, effectively reducing the misjudgment rate. This application also performs target detection through target detection algorithm models such as the YOLO model. After professional labeling and training, it can quickly identify the opening and closing status of subway doors and the risk of people being trapped. Through real-time monitoring and analysis, it has the ability of self-learning and optimization, and can issue alarms in a timely manner to ensure efficient performance in various usage scenarios, further improving the safety and reliability of the public transportation system.

[0094] By combining the RGB images and thermal images obtained by the binocular camera, the system can effectively identify passengers and obstacles in the subway environment, significantly improving the detection accuracy of people-entrapment incidents. The introduction of thermal imaging technology enables the system to clearly distinguish between real people and transparent curtain doors or background objects, especially in low light or complex environments, and still maintain efficient recognition capabilities.

[0095] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the storage medium embodiment and the computer device embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0096] The present application can be used in many general or special computing system environments or configurations, such as personal computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, network PCs, minicomputers, and distributed computing environments including any of the above systems or devices.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not performed.

[0098] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0099] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0100] The above descriptions are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for detecting a person-entrapment accident, characterized in that: include: Acquire an RGB image and a thermal imaging image when the automatic curtain door is closed, and detect a human target at the closing position of the automatic curtain door in the RGB image; In response to detecting a human target at a closed position in the automatic curtain door, marking the human target in the thermal imaging image, and identifying joints of the human target to form a skeleton frame diagram of the human target; Performing temperature sampling on the human target in the thermal imaging image based on the skeleton framework diagram to obtain the temperature of the human target; In response to the temperature of the human target being within a preset body temperature threshold range, a reminder is given of a human trapping accident.

2. The method for detecting a person-entrapment accident according to claim 1, characterized in that: The step of marking the human target in the thermal imaging image and identifying the joints of the human target to form a skeleton frame diagram of the human target includes: Marking the human target in the thermal imaging image aligned with the RGB image by a target detection model; The joints of the human target are identified by a key point detection algorithm to form a skeleton framework diagram of the human target.

3. The method for detecting a person-entrapment accident according to claim 2, characterized in that: The step of performing temperature sampling on the human target in the thermal imaging image based on the skeleton framework diagram to obtain the temperature of the human target includes: Performing temperature sampling on each key bone point in the bone framework diagram; The temperature of the human target is calculated based on the temperature sampling results of each key bone point of the skeleton framework diagram.

4. The method for detecting a person-entrapment accident according to claim 3, characterized in that: The calculating the temperature of the human target based on the temperature sampling results of each key bone point of the skeleton framework diagram includes: Based on the temperature sampling results of each key bone point of the skeleton framework diagram and the corresponding temperature weights, the weighted average temperature of the human target is calculated.

5. The method for detecting a person-entrapment accident according to claim 1, characterized in that: The step of acquiring the RGB image and the thermal imaging image when the automatic curtain door is closed includes: The RGB image and the thermal imaging image are collected at the same viewing angle when the automatic curtain door is closed by a binocular camera; The RGB image and the thermal imaging image are aligned by establishing a mapping relationship between corresponding feature points in the RGB image and the thermal imaging image.

6. The method for detecting a person-entrapment accident according to claim 1, characterized in that: Before detecting the human target at the closed position of the automatic curtain door in the RGB image, the method further includes: Obtain RGB images of automatic curtain doors in different switch states and different lighting environments, and annotate the obtained RGB images to construct training sets and validation sets; Training a deep learning-based target detection algorithm model using the training set; Verifying the performance of the target detection algorithm model using the verification set; The RGB image containing the automatic curtain door is used as the input of the trained object detection algorithm model to detect the open and closed state of the automatic curtain door.

7. The method for detecting a person-entrapment accident according to claim 6, characterized in that: The training of the deep learning-based target detection algorithm model by using the training set includes: Optimizing the parameters of the target detection algorithm model through back propagation and gradient descent algorithms; In response to the loss function constructed by the target detection algorithm model during the back-propagation process converging, the training is terminated.

8. A storage medium having program data stored thereon, characterized in that: When the program data is executed by a processor, the steps of the method for detecting a person-entrapment accident as described in any one of claims 1 to 7 are implemented.

9. A computer device, characterized in that: The method comprises a processor and a memory connected to each other, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for detecting a person-entrapment accident as described in any one of claims 1 to 7 are implemented.

10. A system for detecting a person-entrapment accident, characterized in that: comprising a binocular camera and a computer device as claimed in claim 9; The binocular camera is installed above the automatic curtain door, and is used to continuously acquire RGB images and thermal imaging images and send them to the computer device; The computer device receives the RGB image and the thermal imaging image sent by the binocular camera to execute the human-entrapment accident detection method.

Images