A real-time train oil unloading worker fall identification method
By using high-definition explosion-proof cameras and image processing technology in the oil unloading scenario on trains, the location and characteristics of workers can be automatically identified, enabling real-time monitoring and early warning of employee falls. This solves the problem of security personnel having difficulty monitoring multiple cameras and improves safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-20
- Publication Date
- 2026-03-31
AI Technical Summary
During the unloading of oil from the train, security personnel had difficulty monitoring multiple cameras in real time, which resulted in emergencies such as staff falling not being detected in time, leading to poor safety.
Multiple high-definition explosion-proof cameras were used to capture images of the train unloading oil scene. Human detection models and key point models were used to detect human position bounding boxes, extract features, and identify falls, automatically outputting early warning prompts to avoid security personnel having to check the cameras in real time.
It enables real-time identification and early warning of staff falls, improving safety, eliminating the need for real-time manual monitoring, and reducing safety hazards.
Smart Images
Figure CN117315777B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and in particular to a method, apparatus, equipment, and storage medium for real-time identification of employees falling during oil unloading on trains. Background Technology
[0002] Oil is one of the most popular and supported energy sources today. Oil depot transfer processes include various methods, such as road unloading, train unloading, and pipeline unloading. During train unloading, there is a risk of workers slipping or falling off the train. Currently, security personnel monitor real-time footage from the control room, using their experience to assess the workers' condition and provide warnings and voice prompts based on the footage to prevent accidents and injuries, ensuring the safety of oil depot unloading operations.
[0003] However, the scene of unloading oil by train at the oil depot corresponds to many cameras. Security personnel generally cannot keep an eye on multiple cameras at the same time, nor can they keep up with the situation in real time. They are prone to being distracted and failing to observe the situation of the staff, resulting in poor security. Summary of the Invention
[0004] This invention provides a method, device, equipment, and storage medium for real-time train oil unloading worker fall detection. It extracts images of the work scene of train oil unloading workers captured by multiple cameras, performs person detection and feature extraction, and identifies falls based on the person's position bounding box and classification feature data. It eliminates the need for security personnel to constantly monitor the cameras, ensuring high security.
[0005] In a first aspect, embodiments of the present invention provide a method for real-time train oil unloading employee fall detection, comprising:
[0006] The current frame image captured by the camera is used to detect the bounding box of the person based on the person detection model;
[0007] Anomaly detection is performed on the detected person location bounding boxes;
[0008] Based on the key point model, feature extraction and classification of target people in the person location box are performed to obtain key point data and employee uniform classification data.
[0009] Determine whether the target person corresponding to the employee uniform classification data is an employee;
[0010] Based on the key point data, fall detection is performed on the personnel location boxes corresponding to all employees;
[0011] Output the employee fall detection results;
[0012] Update the position information of the person frame in the current frame image to perform employee fall recognition in the next moment.
[0013] In one embodiment, the anomaly detection of the detected person location box specifically includes:
[0014] When performing drop detection for the first time, it is assumed that no displacement anomaly has occurred.
[0015] When performing fall detection for the second or subsequent times, the first coordinate value of the currently detected person's location box in the current frame image is compared with the second coordinate value of the employee's fall detection person's location box in the previous frame image. If the absolute value of the difference between the first coordinate value and the second coordinate value is greater than a preset value, a warning message is output.
[0016] In one embodiment, the anomaly detection of the detected person location box specifically includes:
[0017] When performing drop detection for the first time, it is assumed that no drop has occurred.
[0018] When performing fall detection for the second or subsequent times, the number of currently detected person location boxes is compared with the number of person location boxes detected in the previous fall detection. If the number of currently detected person location boxes is less than the number of person location boxes detected in the previous fall detection, a warning message is output.
[0019] In one embodiment, the step of extracting and classifying features of the target person in the person location box based on the key point model to obtain key point data and employee uniform classification data specifically includes:
[0020] Extract the clothing data of the target person corresponding to the person's location box to obtain employee uniform classification data.
[0021] In one embodiment, the step of extracting and classifying features of the target person in the person location box based on the key point model to obtain key point data and employee uniform classification data specifically includes:
[0022] Extract the body shape data of the target person corresponding to the person's position box to obtain key point data. The key point data includes the shoulder position and the hip position, and the key point data includes the line connecting the shoulders and the line connecting the hips.
[0023] In one embodiment, determining whether the target person corresponding to the employee uniform classification data is an employee specifically includes:
[0024] The system retrieves clothing data and compares it with clothing data in a pre-defined employee database. If the comparison is successful, the system determines that the target person corresponding to the employee clothing category data is an employee. If the comparison fails, the system outputs a warning message.
[0025] In one embodiment, the step of performing fall detection on the bounding boxes of all employees based on the key point data specifically includes:
[0026] Determine whether the plane formed by the line connecting the two shoulders and the line connecting the two spans is parallel to the ground. If it is parallel, output a warning message.
[0027] The system compares the time elapsed when the plane formed by the lines connecting the two shoulders and the two spans is parallel to the ground with a preset time. If the preset time is exceeded, a warning message is output.
[0028] Secondly, embodiments of the present invention provide an identification device applied to the real-time train oil unloading employee fall detection method as described in the first aspect, comprising an acquisition and detection module, a judgment and identification module, an extraction and classification module, an output module, and an update module; wherein...
[0029] The acquisition and detection module is used to acquire the current frame image captured by the camera and perform person location bounding box detection based on the person detection model;
[0030] The judgment and recognition module is used to identify anomalies in the detected person location boxes, and is also used to determine whether the target person corresponding to the employee uniform classification data is an employee after extraction and classification, and to perform fall recognition on the person location boxes corresponding to all employees based on the key point data.
[0031] The extraction and classification module is used to extract and classify the features of the target person in the person location box based on the key point model, so as to obtain key point data and employee uniform classification data.
[0032] The output module is used to output the employee fall detection results;
[0033] The update module is used to update the position information of the person frame in the current frame image and to perform employee fall recognition in the next moment.
[0034] Thirdly, embodiments of the present invention provide an identification device, a memory, and a processor, wherein the processor is connected to the memory; wherein...
[0035] The memory is used to store program instructions;
[0036] The processor is used to invoke program instructions in the memory to execute some or all of the steps included in the real-time train unloading employee fall detection method as described in the first aspect.
[0037] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform some or all of the steps of the real-time train unloading employee fall detection method described in the first aspect.
[0038] In this embodiment of the invention, a person location bounding box is detected based on a person detection model by acquiring the current frame image captured by a camera; anomaly identification is performed on the detected person location bounding boxes; feature extraction and classification of the target person in the person location bounding box are performed based on a key point model to obtain key point data and employee uniform classification data; it is determined whether the target person corresponding to the employee uniform classification data is an employee; fall detection is performed on the person location bounding boxes corresponding to all employees based on the key point data; and the employee fall detection result is output. Through person target detection and person feature key point fusion, person detection and feature extraction are achieved in images of staff working scenes, and fall detection is identified, outputting early warning information. This eliminates the need for security personnel to constantly monitor the camera, resulting in high security. Attached Figure Description
[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0040] Figure 1 This is a flowchart illustrating the real-time train oil unloading employee fall detection method provided in this embodiment of the invention.
[0041] Figure 2 This is a schematic diagram illustrating the specific process of the real-time train oil unloading employee fall detection method provided in this embodiment of the invention.
[0042] Figure 3 This is a key point model architecture diagram provided in the embodiments of the present invention.
[0043] Figure 4 This is a schematic diagram of the structure of the identification device provided in an embodiment of the present invention.
[0044] Figure 5 This is a schematic diagram of the structure of the identification device provided in an embodiment of the present invention.
[0045] In the diagram: 20-Identification device, 201-Acquisition and detection module, 202-Judgment and identification module, 203-Extraction and classification module, 204-Output module, 205-Update module, 400-Identification device, 401-Processor, 402-Memory, 403-Communication interface, 404-Bus. Detailed Implementation
[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0047] Please see Figure 1 , Figure 1This is a flowchart illustrating a real-time train oil unloading employee fall detection method provided by the present invention. Specifically, as shown... Figure 1 As shown, the real-time train oil unloading employee fall detection method may include the following steps:
[0048] S101. Obtain the current frame image captured by the camera and perform person location bounding box detection based on the person detection model.
[0049] In this embodiment of the invention, multiple high-definition explosion-proof cameras are installed within the oil unloading area of the train. These cameras are at least 2 megapixels (1920*1080), and the distance between the camera and the deployment area is less than 10 meters and greater than 1 meter. They are IPx6 waterproof. Specifically, the camera footage of the oil depot's train unloading scene is captured via streaming media. One frame of the streaming media image is used as input to a person detection model. The model detects multiple person bounding boxes in the camera image. Before detecting the person bounding boxes, the method further includes data augmentation processing such as flipping, adjusting brightness, noise, and mosaic effects on the current frame image; and learning rate decay and multi-scale training are applied to the person detection model.
[0050] S102. Perform anomaly detection on the detected person location box.
[0051] In this embodiment of the invention, all person position boxes are compared with the previously identified set of person position boxes to check for significant displacement or abnormal changes in the number of person boxes. This excludes cases where the person position boxes disappear at the edge of the image. For the first identification, the previous task position box set is empty, indicating no fall or abnormal displacement has occurred. Specifically, for the first fall identification, no abnormal displacement is assumed. For subsequent fall identifications, the first coordinate value of the currently detected person position box in the current frame is compared with the second coordinate value of the employee's fall identification person position box in the previous frame. If the absolute value of the difference between the first and second coordinate values is greater than a preset value, a warning message is output; otherwise, key point detection confirms a fall or prolonged lying down identification. For example, if the coordinates of the previously detected person's location frame are and the coordinates of the currently detected person's location frame are and , the default value is Q, which is the absolute value of the difference between them. When Q = 1m, then = 2, and 2 is greater than 1, it means that the person corresponding to the location frame is more than 2 meters away from the previous position, which exceeds the normal oil unloading operation distance. An over-distance warning message is output to prompt security personnel to check for potential safety accidents. When Q = 1m, then = 0.9, and 0.9 is less than or greater than 1, it means that the person corresponding to the location frame is 0.9 meters away from the previous position, which is within the normal oil unloading operation movement distance and does not require processing. When this is the first fall detection, it is assumed that no fall has occurred. When this is the second or subsequent fall detection, the number of currently detected person's location frames is compared with the number of person's location frames detected in the previous fall detection. If the number of currently detected person's location frames is less than the number of person's location frames detected in the previous fall detection, a warning message is output. For example, if the number of person location frames detected in the previous moment was 6 and the number of person location frames detected in the current moment is 5, an abnormal number of people warning information will be output to prompt safety personnel to check whether a fall accident has occurred or whether people are resting, so as to facilitate management and improve safety.
[0052] S103. Based on the key point model, perform feature extraction and classification on the target person in the person location box to obtain key point data and employee uniform classification data.
[0053] In the embodiments of this invention, please refer to Figure 3 , Figure 3This is a key point model architecture diagram provided in an embodiment of the present invention. Clothing data of the target person corresponding to the person's location box is extracted to obtain employee uniform classification data. The clothing data includes clothing color, clothing graphic, clothing logo, and clothing style. Clothing colors include one or more of red, yellow, and blue. Clothing graphics include a hat pattern on the back of the clothing. Clothing logos include a company logo on the front of the clothing. Clothing styles include pockets on the side of the pants. Body shape data of the target person corresponding to the person's location box is extracted to obtain key point data. The key point data includes shoulder positions and hip positions, and includes lines connecting the shoulders and hips.
[0054] S104. Determine whether the target person corresponding to the employee uniform classification data is an employee.
[0055] In this embodiment of the invention, clothing data is obtained and compared with clothing data in a preset employee database. If the comparison is successful, the target person corresponding to the employee clothing category data is determined to be an employee. If the comparison fails, an early warning prompt is output.
[0056] In this embodiment of the invention, the clothing color, clothing graphic, clothing logo, and clothing style of the target person obtained from the extracted person location box are compared one by one with the clothing color, clothing graphic, clothing logo, and clothing style of the preset employee database. If any one of them fails the comparison, a non-employee intrusion warning is output.
[0057] The method further includes: when it is determined that the target person corresponding to the employee uniform classification data is an employee, the facial image of the target employee is obtained and compared with the facial images of employees in the preset shift schedule database. If the comparison fails, an early warning message is output, indicating that the employee is not on duty, prompting the management personnel to pay attention to whether employees are arbitrarily adjusting their work shifts and engaging in private activities during unreported work hours, so as to facilitate the management personnel to manage accordingly.
[0058] The method further includes: when the target employee's facial image fails to match the employee facial images in the preset shift schedule database, obtaining the target employee's facial image and comparing it with the employee facial images in the preset employee database; if the comparison fails, an early warning message is output; if the comparison succeeds, an abnormal personnel message is output. This step avoids having other personnel substitute for the company's trained oil unloading employees to perform oil unloading operations, preventing safety accidents due to professional limitations and improving safety.
[0059] S105. Based on the key point data, perform fall detection on the person position boxes corresponding to all employees and output the employee fall detection results.
[0060] In this embodiment of the invention, it is determined whether the plane formed by the line connecting the shoulders and the line connecting the hips is parallel to the ground. If it is parallel, a warning message is output, indicating that the employee is in a lying position, and a lying-down warning message is output to remind security personnel to pay attention. The time when the plane formed by the line connecting the shoulders and the line connecting the hips is parallel to the ground is compared with a preset time. If the preset time is exceeded, a warning message is output. Through this step, exceeding the preset time indicates that the employee has been lying flat for a long time and the fall has lasted for a long time, requiring notification of medical personnel for treatment, shortening the emergency response time, and improving the chances of survival. In addition to the above judgment method, the application can be adapted according to the specific scene camera. For example, according to the numerical order of the skeletal key points, corresponding to the skeletal position, when the key points of the shoulders are perpendicular to the key points of the hips, it can be considered as a lying position, or when they are in a straight line.
[0061] S106. Update the position information of the person frame in the current frame image and perform employee fall recognition in the next moment.
[0062] In this embodiment of the invention, an early warning is issued when a fall or prolonged lying down occurs; otherwise, the position of the character frame in the current deduction is updated to the previous position frame information for the next judgment. Similarly, after an early warning, the position of the character frame in the current deduction is updated to the previous position frame information for the next judgment.
[0063] In the embodiments of this invention, please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the specific process of the real-time train oil unloading employee fall detection method provided in this embodiment of the invention. It involves acquiring the current frame image captured by a camera and performing person bounding box detection based on a person detection model; identifying anomalies in the detected bounding boxes; extracting and classifying features of the target person within the bounding box based on a keypoint model to obtain keypoint data and employee uniform classification data; determining whether the target person corresponding to the employee uniform classification data is an employee; performing fall detection on the bounding boxes corresponding to all employees based on the keypoint data; and outputting the employee fall detection result. Through person target detection and keypoint fusion, it achieves person detection and feature extraction in images of workers' work scenes, identifies whether a fall has occurred, and outputs warning messages. This eliminates the need for security personnel to constantly monitor the cameras, ensuring high security.
[0064] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an identification device 20 provided in an embodiment of the present invention. The identification device 20 of this embodiment includes a module for performing the above-described real-time train oil unloading employee fall identification method. Specifically, the identification device 20 of this embodiment may include: an acquisition and detection module 201, a judgment and identification module 202, an extraction and classification module 203, an output module 204, and an update module 205. Wherein:
[0065] The acquisition and detection module 201 is used to acquire the current frame image captured by the camera and perform person position box detection based on the person detection model;
[0066] The judgment and recognition module 202 is used to perform anomaly recognition on the detected person location box, and is also used to determine whether the target person corresponding to the employee uniform classification data is an employee after extraction and classification, and to perform fall recognition on the person location box corresponding to all employees based on the key point data.
[0067] The extraction and classification module 203 is used to extract and classify the features of the target person in the person location box based on the key point model, so as to obtain key point data and employee uniform classification data.
[0068] The output module 204 is used to output the employee fall recognition results;
[0069] The update module 205 is used to update the position information of the person frame in the current frame image and to perform employee fall recognition in the next moment.
[0070] In one embodiment, the judgment and identification module 202 is specifically used for:
[0071] When performing drop detection for the first time, it is assumed that no drop has occurred.
[0072] When performing fall detection for the second or subsequent times, the number of currently detected person location boxes is compared with the number of person location boxes detected in the previous fall detection. If the number of currently detected person location boxes is less than the number of person location boxes detected in the previous fall detection, a warning message is output.
[0073] In one embodiment, the extraction and classification module 203 is specifically used for:
[0074] Extract the clothing data of the target person corresponding to the person's location box to obtain employee uniform classification data.
[0075] In one embodiment, the extraction and classification module 203 is specifically used for:
[0076] Extract the body shape data of the target person corresponding to the person's position box to obtain key point data. The key point data includes the shoulder position and the hip position, and the key point data includes the line connecting the shoulders and the line connecting the hips.
[0077] In one embodiment, the judgment and identification module 202 is specifically used for:
[0078] The system retrieves clothing data and compares it with clothing data in a pre-defined employee database. If the comparison is successful, the system determines that the target person corresponding to the employee clothing category data is an employee. If the comparison fails, the system outputs a warning message.
[0079] In one embodiment, the judgment and identification module 202 is specifically used for:
[0080] Determine whether the plane formed by the line connecting the two shoulders and the line connecting the two spans is parallel to the ground. If it is parallel, output a warning message.
[0081] The system compares the time elapsed when the plane formed by the lines connecting the two shoulders and the two spans is parallel to the ground with a preset time. If the preset time is exceeded, a warning message is output.
[0082] Please participate Figure 5 , Figure 5 This is a schematic diagram of the structure of an identification device 400 provided in an embodiment of the present invention. The identification device is used to perform the aforementioned real-time train oil unloading employee fall detection method. Figure 5 The identification device 400 in this embodiment may include one or more processors 401 and a memory 402. Optionally, the identification device 400 may also include one or more communication interfaces 403. The processor 401, communication interface 403, and memory 402 can be connected via a bus 404, or via other means. Figure 5 The example is illustrated using a bus-based approach.
[0083] The processor 401 can be a Central Processing Unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.
[0084] The communication interface 403 can be used for exchanging information or signaling, as well as receiving and transmitting signals. The communication interface 403 may include a receiver and a transmitter for communicating with other devices. The memory 402 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one stored program required for a function (such as text storage function, location storage function, etc.). The data storage area may store data created based on the use of the recognition device (such as image data, text data), and may include application stored programs. Furthermore, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0085] The memory 402 is also used to store program instructions. The processor 401 can call the program instructions stored in the memory 402 to implement the real-time train oil unloading employee fall detection method as shown in the embodiment of the present invention.
[0086] In specific implementations, the processor 401, etc., described in the embodiments of the present invention can execute the above-mentioned... Figure 1 or Figure 2 The implementation methods described in the illustrated embodiments can also be used to execute the embodiments of the present invention. Figure 4 The implementation methods of each module are not detailed here.
[0087] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can perform... Figure 1 or Figure 2 Some or all of the steps in the real-time train unloading employee fall detection method described in the corresponding embodiments can also be used to implement the present invention. Figure 4 The function of the identification device 20 in the illustrated embodiment can also realize the present invention. Figure 5 The functions of the identification device 400 in the illustrated embodiment will not be described in detail here.
[0088] The computer-readable storage medium may be the internal storage module of the real-time train unloading employee fall detection method or device described in the foregoing embodiments, such as the hard drive or memory of the detection device or device. The computer-readable storage medium may also be an external storage device of the detection device or device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the detection device or device.
[0089] This invention also provides a computer program product containing instructions that, when run on a computer, causes the computer to perform some or all of the steps in the above-described method.
[0090] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.
[0091] In the various embodiments of this application, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this invention.
[0092] The above description is only a partial embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.
Claims
1. A real-time train oiler employee fall identification method, characterized by, The method comprises the following steps: detecting a person position frame of a current frame picture captured by a camera based on a person detection model; performing abnormality identification on the detected person position frame; performing feature extraction and classification on a target person in the person position frame based on a key point model to obtain key point data and employee uniform classification data; determining whether the target person corresponding to the employee uniform classification data is an employee; performing fall identification on the person position frames corresponding to all employees based on the key point data; outputting employee fall identification results; updating person frame position information of the current frame picture to perform employee fall identification at a next time; wherein the abnormality identification on the detected person position frame specifically comprises: when the fall identification is performed for the first time, it is determined by default that no displacement abnormality occurs; when the fall identification is performed for the second time and above, a first coordinate value of the current detected person position frame in the current frame picture is compared with a second coordinate value of the employee fall identification person position frame in the last frame picture at a last time, and if an absolute value of a difference between the first coordinate value and the second coordinate value is greater than a preset value, a warning prompt information is outputted; when the fall identification is performed for the first time, it is determined by default that no fall occurs; when the fall identification is performed for the second time and above, a number of the current detected person position frames is compared with a number of the last fall identification person position frame, and if the number of the current detected person position frames is less than the number of the last fall identification person position frame, the warning prompt information is outputted.
2. The real-time train oiler crew fall identification method of claim 1, wherein, The feature extraction and classification on the target person in the person position frame based on the key point model to obtain the key point data and the employee uniform classification data specifically comprises: extracting clothing data of the target person corresponding to the person position frame to obtain the employee uniform classification data.
3. The real-time train oiler crew fall identification method of claim 2, wherein, The feature extraction and classification on the target person in the person position frame based on the key point model to obtain the key point data and the employee uniform classification data specifically comprises: extracting body shape data of the target person corresponding to the person position frame to obtain the key point data, wherein the key point data comprises double-shoulder positions and double-cross positions, and the key point data comprises a double-shoulder line and a double-cross line.
4. The real-time train oiler crew fall identification method of claim 3, wherein, The determination of whether the target person corresponding to the employee uniform classification data is an employee specifically comprises: comparing the clothing data with clothing data of a preset employee database, and if the comparison is qualified, it is determined that the target person corresponding to the employee uniform classification data is an employee, and if the comparison is unqualified, a warning prompt is outputted.
5. The real-time train oiler crew fall identification method of claim 4, wherein, The fall identification on the person position frames corresponding to all employees based on the key point data specifically comprises: determining whether a plane formed by the double-shoulder line and the double-cross line is parallel to the ground, and if so, a warning prompt information is outputted; comparing a time when the plane formed by the double-shoulder line and the double-cross line is parallel to the ground with a preset time, and if the time exceeds the preset time, the warning prompt information is outputted.
6. A recognition device applied to the real-time train oil unloading staff falling recognition method according to any one of claims 1 to 5, characterized in that, The method comprises an acquisition detection module, a determination and identification module, an extraction and classification module, an output module and an update module, wherein: the acquisition detection module is configured to detect a person position frame of a current frame picture captured by a camera based on a person detection model; The judgment and recognition module is configured to perform abnormal recognition on the detected human position box, and is further configured to judge whether a target human corresponding to the employee uniform classification data is an employee after extracting and classifying, and perform fall recognition on human position boxes corresponding to all employees based on the key point data; The extraction and classification module is configured to perform feature extraction and classification on a target human of a human position box based on a key point model to obtain key point data and employee uniform classification data; The output module is configured to output an employee fall recognition result. The update module is configured to update human box position information of a current frame of picture for employee fall recognition at a next time.
7. An identification device, characterized in that Comprise: A memory and a processor connected with the memory; wherein, The memory is configured to store program instructions; The processor is configured to call the program instructions in the memory to execute part or all of the steps of the real-time train oil unloading employee fall recognition method according to any one of claims 1 to 5. 8.A computer readable storage medium, the computer storage medium storing a computer program, the computer program comprising program instructions, the program instructions causing a processor to execute part or all of the steps of the real-time train oil unloading employee fall recognition method according to any one of claims 1 to 5 when the program instructions are executed by the processor.
Citation Information
Patent Citations
Falling behaviour detection method and device
WO2021227874A1