AI abnormal behavior identification and early warning system for construction site safety supervision

Abstract

This invention belongs to the field of safety monitoring technology, specifically disclosing an AI-based abnormal behavior recognition and early warning system for construction site safety supervision. Through an intelligent visual safety monitoring system comprising a monitoring server, data communication equipment, alarm terminals, and several monitoring cameras, it can collect monitoring images of personnel within corresponding work scenarios on the construction site. Based on the collected personnel monitoring images, it performs target key point detection and calculates key point spatial feature parameters to determine whether personnel in the work scenario exhibit abnormal behavior, thereby achieving targeted safety early warnings. This can improve the accuracy and real-time performance of abnormal behavior recognition in complex construction site scenarios, significantly reducing the false alarm rate. This invention can automatically achieve full coverage and uninterrupted long-term monitoring of abnormal behavior of construction site workers, saving inspection manpower, greatly improving the level of construction site safety management, and effectively reducing the incidence of construction site safety accidents.

Description

Technical Field

[0001] This invention belongs to the field of safety monitoring technology, specifically relating to an AI-based abnormal behavior recognition and early warning system for construction site safety supervision. Background Technology

[0002] The construction industry, as a pillar industry of the national economy, is also one of the high-risk industries. Currently, the safety situation in the construction industry remains quite serious. Accidents such as falls from heights, falling objects, and machinery injuries caused by improper work practices occur frequently, resulting in significant losses of life and property for construction workers. Traditional manual safety inspection methods at construction sites suffer from limited coverage time and scope, delayed response, and subjective judgment bias, making them unsuitable for the needs of modern construction site safety management. Summary of the Invention

[0003] The purpose of this invention is to provide an AI-based abnormal behavior recognition and early warning system for construction site safety supervision, in order to solve the aforementioned problems existing in the prior art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides an AI-based abnormal behavior recognition and early warning system for construction site safety supervision, comprising a monitoring server, data communication equipment, an alarm terminal, and several monitoring cameras, wherein: Surveillance cameras are deployed at the target work scene on the construction site to collect real-time personnel monitoring images of the target work scene, obtain personnel monitoring images at various times, and use the personnel monitoring images at various times to form a personnel monitoring image sequence, and transmit the personnel monitoring image sequence to the data communication equipment; Data communication equipment is used to upload the sequence of personnel monitoring images transmitted by surveillance cameras to the monitoring server, and to transmit the security warning information issued by the monitoring server to the alarm terminal. The alarm terminal is deployed in the target work scene of the construction site to receive safety warning information transmitted by data communication equipment and to play and display the safety warning information. The monitoring server receives personnel monitoring image sequences uploaded by data communication equipment, performs target detection on each personnel monitoring image in the sequence to identify the target personnel in each image, performs key point detection on the target personnel in the images to obtain several pose key points of the target personnel, determines the spatial feature parameters of the target personnel's behavior based on the several pose key points, determines whether the target personnel's behavior is abnormal based on the spatial feature parameters of the target personnel's behavior, and generates safety warning information when the target personnel's behavior is determined to be abnormal, and transmits the safety warning information to the data communication equipment.

[0005] In one possible design, the monitoring server performs target detection on each personnel monitoring image in the personnel monitoring image sequence, identifies the target personnel in each personnel monitoring image, performs key point detection on the target personnel in the personnel monitoring images, and obtains several pose key points of the target personnel. Then, it inputs each personnel monitoring image in the personnel monitoring image sequence into a pre-trained YOLOv11-Pose model for target detection, determines the bounding box of the target personnel in each personnel monitoring image, and performs human pose estimation on the target personnel within the bounding box of the target personnel through the pre-trained YOLOv11-Pose model to obtain several pose key points of the target personnel.

[0006] In one possible design, the monitoring server determines the spatial feature parameters of the target person's behavior based on several posture key points. When determining whether the target person's behavior is abnormal based on these parameters, the server selects the hip posture key point as the root node from all posture key points, determines the direction vector of the root node relative to the other posture key points, and calculates the planar angle θ between these direction vectors. ,in, and These are two different direction vectors. The plane angles of each vector are then used to form the plane angle feature vector of the target person. The plane angle feature vector of the target person is then input into a pre-trained action recognition model for action recognition, and the corresponding behavior action classification results are obtained. Based on the behavior action classification results, it is determined whether the target person's behavior action is abnormal.

[0007] In one possible design, the monitoring server determines the spatial feature parameters of the target person's behavior based on several posture key points. When determining whether the target person's behavior is abnormal based on the spatial feature parameters, the hip posture key point is selected as the root node from all posture key points, and a convex hull of all posture key points is constructed. The circumcircle radius R of the convex hull and the distance from the root node to each side of the convex hull are determined. The shortest distance from the root node to each side of the convex hull is taken as the feature distance L. The stability coefficient Q of the target person is calculated using the feature distance L and the circumcircle radius R, where Q = L / R. When the stability coefficient Q is less than the set stability coefficient threshold, the target person's behavior is determined to be abnormal.

[0008] In one possible design, the monitoring server determines the spatial feature parameters of the target person's behavior based on several posture key points. When determining whether the target person's behavior is abnormal based on these spatial feature parameters, it selects three target key points from all posture key points and uses these three key points to construct the target joint angles of the target person. The target joint angles of the target person on the monitoring images corresponding to three consecutive time points are then substituted into a preset joint angular momentum feature calculation model. The calculation is performed to obtain the joint angular momentum characteristic value M of the target person. When the joint angular momentum characteristic value M of the target person exceeds the set angular momentum threshold, the target person's behavior is determined to be abnormal. , and These are the target joint angles of the target personnel in the personnel monitoring images at three consecutive time points, where Δt is the time interval between each time point, and I is the set equivalent inertia coefficient.

[0009] In one possible design, the monitoring server is further configured to mark the warning area in the corresponding personnel monitoring image according to the preset warning area image coordinate parameters, and determine whether the key points of the target personnel's posture are within the marked warning area. When it is determined that the key points of the target personnel's posture are within the marked warning area, the server generates area intrusion warning information and transmits the area intrusion warning information to the data communication device. The data communication device is further configured to transmit the area intrusion warning information issued by the monitoring server to the alarm terminal. The alarm terminal is further configured to receive the area intrusion warning information transmitted by the data communication device and display the area intrusion warning information.

[0010] In one possible design, the monitoring server is also used to detect whether a target person in the personnel monitoring image is wearing a safety helmet, and to send a safety warning message to the data communication device when it detects that the target person in the personnel monitoring image is not wearing a safety helmet.

[0011] In one possible design, when the monitoring server performs helmet-wearing detection on a target person in a personnel monitoring image, it extracts the target person image region from the personnel monitoring image, performs image enhancement and grayscale processing on the target person image region to obtain a grayscale image of the target person, and uses a pre-trained YOLOv11 model to perform helmet detection on the grayscale image of the target person to obtain the corresponding helmet-wearing detection result.

[0012] In one possible design, the alarm terminal includes a voice alarm and a light alarm, wherein the voice alarm is used to issue a warning voice based on the safety warning information, and the light alarm is used to issue a warning light based on the safety warning information.

[0013] Beneficial Effects: This invention uses an intelligent visual safety monitoring system to collect images of personnel in corresponding work scenarios on construction sites. Based on the collected personnel monitoring images, it performs target key point detection and calculates key point spatial feature parameters to determine whether personnel in the work scenario exhibit abnormal behavior, thereby achieving targeted safety warnings. This can improve the accuracy and real-time performance of abnormal personnel behavior identification in complex construction site scenarios and significantly reduce the false alarm rate. This invention can automatically achieve full coverage and uninterrupted long-term monitoring of abnormal behavior of construction site workers, saving inspection manpower, greatly improving the level of construction site safety management, and effectively reducing the incidence of construction site safety accidents. Attached Figure Description

[0014] 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.

[0015] Figure 1 This is a schematic diagram of the system configuration of the present invention; Figure 2 This is a schematic diagram of the monitoring server's structure. Detailed Implementation

[0016] It should be noted that the descriptions of these embodiments are intended to aid in understanding the invention and do not constitute a limitation thereof. The specific structural and functional details disclosed herein are merely for describing exemplary embodiments of the invention. However, the invention may be embodied in many alternative forms and should not be construed as being limited to the embodiments described herein.

[0017] It should be understood that, unless otherwise explicitly specified and limited, the corresponding terms should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be an electrical connection, a direct connection, or an indirect connection through an intermediate medium; it can also refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments according to the specific circumstances.

[0018] Specific details are provided in the following description to provide a complete understanding of the exemplary embodiments. However, those skilled in the art will understand that the exemplary embodiments can be implemented without these specific details. For example, the system may be shown in block diagrams to avoid obscuring the example with unnecessary details. In other embodiments, well-known processes, structures, and techniques may be shown without non-essential details to avoid obscuring the embodiments.

[0019] Example: This embodiment provides an AI-based abnormal behavior recognition and early warning system for construction site safety supervision, such as... Figure 1 As shown, it includes a monitoring server, data communication equipment, alarm terminal, and several surveillance cameras, among which: Surveillance cameras are deployed at the target work scene on the construction site to collect real-time personnel monitoring images of the target work scene, obtain personnel monitoring images at various times, and use the personnel monitoring images at various times to form a personnel monitoring image sequence, and transmit the personnel monitoring image sequence to the data communication equipment; Data communication equipment is used to upload the sequence of personnel monitoring images transmitted by surveillance cameras to the monitoring server, and to transmit the security warning information issued by the monitoring server to the alarm terminal. The alarm terminal is deployed in the target work scene of the construction site to receive safety warning information transmitted by data communication equipment and to play and display the safety warning information. The monitoring server receives personnel monitoring image sequences uploaded by data communication equipment, performs target detection on each personnel monitoring image in the sequence to identify the target personnel in each image, performs key point detection on the target personnel in the images to obtain several pose key points of the target personnel, determines the spatial feature parameters of the target personnel's behavior based on the several pose key points, determines whether the target personnel's behavior is abnormal based on the spatial feature parameters of the target personnel's behavior, and generates safety warning information when the target personnel's behavior is determined to be abnormal, and transmits the safety warning information to the data communication equipment.

[0020] In practice, surveillance cameras deployed at the construction site can capture real-time images of personnel in the target work area. These images are then compiled into a sequence and transmitted to a data communication device, which in turn uploads them to a monitoring server. The monitoring server performs target detection and key point detection on each image in the sequence to identify key pose points of the corresponding personnel. Based on these key pose points, spatial feature parameters of the personnel's actions are extracted. These parameters are then used to determine if the personnel's actions are abnormal. If abnormality is detected, a safety warning is generated and sent to the data communication device, which then forwards it to an alarm terminal. Finally, the alarm terminal deployed at the construction site displays the safety warning to alert workers in the target work area to adhere to proper work procedures and prevent accidents.

[0021] When the monitoring server performs target detection and keypoint detection on each person monitoring image in a personnel monitoring image sequence, it can input each person monitoring image in the sequence into a pre-trained YOLOv11-Pose model for target detection, determine the bounding box of the target person in each monitoring image, and then use the pre-trained YOLOv11-Pose model to estimate the human pose of the target person within the bounding box, obtaining several pose keypoints of the target person. When a monitoring server determines the spatial feature parameters of a target person's behavior based on several key pose points, and then determines whether the target person's behavior is abnormal based on these spatial feature parameters, it can do so in the following ways: 1. Select the hip posture keypoint (the keypoint at the center of gravity) from all posture keypoints as the root node, determine the direction vector of the root node relative to the other posture keypoints, and calculate the plane angle θ between each direction vector. ,in, and These are two different direction vectors. Then, the plane angles are used to form the plane angle feature vector of the target person (to describe joint connectivity and posture structure). The plane angle feature vector of the target person is input into the pre-trained action recognition model for action recognition, and the corresponding behavior action classification result is obtained. Based on the behavior action classification result, it is determined whether the behavior action of the target person is abnormal.

[0022] 2. Select the hip posture key point from all posture key points as the root node, and construct the convex hull of all posture key points. Determine the circumcircle radius R of the convex hull, as well as the distance from the root node to each side of the convex hull. Take the shortest distance from the root node to each side of the convex hull as the feature distance L. Calculate the stability coefficient Q (describing the center of gravity stability) of the target person using the feature distance L and the circumcircle radius R. Q = L / R. When the stability coefficient Q is less than the set stability coefficient threshold, the target person's behavior is judged to be abnormal.

[0023] 3. Select three target key points from all posture key points (such as key points at the shoulder, elbow, and wrist), and use these three target key points to construct the target joint angles of the target person. Substitute the target joint angles of the target person in the monitoring images corresponding to three consecutive time points into a preset joint angular momentum feature calculation model. The calculations are performed to obtain the joint angular momentum characteristic value M of the target person (describing the amplitude of motion and dynamic characteristics). When the joint angular momentum characteristic value M of the target person exceeds the set angular momentum threshold, the target person's behavior is determined to be abnormal. , and These are the target joint angles of the target personnel in the personnel monitoring images at three consecutive time points, where Δt is the time interval between each time point, and I is the set equivalent inertia coefficient.

[0024] Furthermore, the monitoring server is also used to mark the warning area in the corresponding personnel monitoring image according to the preset warning area image coordinate parameters, and to determine whether the key points of the target personnel's posture are within the marked warning area. When it is determined that the key points of the target personnel's posture are within the marked warning area, an area intrusion warning information is generated and transmitted to the data communication device. The data communication device is also used to transmit the area intrusion warning information issued by the monitoring server to the alarm terminal. The alarm terminal is also used to receive the area intrusion warning information transmitted by the data communication device and to play and display the area intrusion warning information. Alternatively, electronic fence technology can be used for monitoring intrusion warnings in the monitored area. That is, electronic fence hardware devices are deployed in the corresponding monitored area of ​​the construction site to monitor intrusion of construction site personnel through the electronic fence, and when it is determined that a target personnel has intruded into the monitored area, an area intrusion warning information is sent to the alarm terminal.

[0025] Furthermore, the monitoring server is also used to detect helmet wearing in target personnel in the personnel monitoring images, and to send a safety warning to the data communication device when it detects that a target personnel in the personnel monitoring images is not wearing a helmet. When detecting helmet wearing in target personnel in the personnel monitoring images, the monitoring server can first extract the target personnel image region from the personnel monitoring images, perform image enhancement and grayscale processing on the target personnel image region to obtain a grayscale image of the target personnel, and then use a pre-trained YOLOv11 model to detect helmet wearing in the grayscale image of the target personnel to obtain the corresponding helmet wearing detection result.

[0026] The surveillance camera may be a high-definition camera with night vision capability. The data communication equipment may employ wireless or wired transmission methods. The alarm terminal may include a voice alarm and a light alarm; the voice alarm is used to issue a warning voice based on safety warning information, and the light alarm is used to issue a warning light based on safety warning information. Figure 2As shown, the monitoring server, at the hardware level, may include a data interface for establishing data communication between the processor and external data terminals, a memory for storing instructions, and a processor for reading instructions stored in the memory and executing them accordingly. Optionally, the monitoring server may also include an internal bus, through which the processor, memory, and data interface can be interconnected. This internal bus may be a PCIe (Peripheral Component Interconnect Express) bus, which may be divided into an address bus, a data bus, a control bus, etc. The memory may include, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Flash Memory, First Input First Output (FIFO), and / or First In Last Out (FILO). The processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), or a Field-Programmable Gate Array (FPGA). Gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0027] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-based abnormal behavior recognition and early warning system for construction site safety supervision, characterized in that: It includes a monitoring server, data communication equipment, alarm terminals, and several surveillance cameras, among which: Surveillance cameras are deployed at the target work scene on the construction site to collect real-time personnel monitoring images of the target work scene, obtain personnel monitoring images at various times, and use the personnel monitoring images at various times to form a personnel monitoring image sequence, and transmit the personnel monitoring image sequence to the data communication equipment; Data communication equipment is used to upload the sequence of personnel monitoring images transmitted by surveillance cameras to the monitoring server, and to transmit the security warning information issued by the monitoring server to the alarm terminal. The alarm terminal is deployed in the target work scene of the construction site to receive safety warning information transmitted by data communication equipment and to play and display the safety warning information. The monitoring server receives personnel monitoring image sequences uploaded by data communication equipment, performs target detection on each personnel monitoring image in the sequence to identify the target personnel in each image, performs key point detection on the target personnel in the images to obtain several pose key points of the target personnel, determines the spatial feature parameters of the target personnel's behavior based on the several pose key points, determines whether the target personnel's behavior is abnormal based on the spatial feature parameters of the target personnel's behavior, and generates safety warning information when the target personnel's behavior is determined to be abnormal, and transmits the safety warning information to the data communication equipment.

2. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 1, characterized in that, The monitoring server performs target detection on each personnel monitoring image in the personnel monitoring image sequence, identifies the target personnel in each personnel monitoring image, performs key point detection on the target personnel in the personnel monitoring image, and obtains several pose key points of the target personnel. Then, it inputs each personnel monitoring image in the personnel monitoring image sequence into a pre-trained YOLOv11-Pose model for target detection, determines the bounding box of the target personnel in each personnel monitoring image, and performs human pose estimation on the target personnel within the bounding box of the target personnel through the pre-trained YOLOv11-Pose model to obtain several pose key points of the target personnel.

3. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 1, characterized in that, The monitoring server determines the spatial feature parameters of the target person's behavior based on several posture key points. When determining whether the target person's behavior is abnormal based on these parameters, the server selects the hip posture key point as the root node from all posture key points, determines the direction vector of the root node relative to the other posture key points, and calculates the plane angle θ between each direction vector. ,in, and These are two different direction vectors. The plane angles of each vector are then used to form the plane angle feature vector of the target person. The plane angle feature vector of the target person is then input into a pre-trained action recognition model for action recognition, and the corresponding behavior action classification results are obtained. Based on the behavior action classification results, it is determined whether the target person's behavior action is abnormal.

4. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 1, characterized in that, The monitoring server determines the spatial feature parameters of the target person's behavior based on several posture key points. When determining whether the target person's behavior is abnormal based on the spatial feature parameters, the hip posture key point is selected as the root node from all posture key points, and the convex hull of all posture key points is constructed. The circumcircle radius R of the convex hull and the distance from the root node to each side of the convex hull are determined. The shortest distance from the root node to each side of the convex hull is taken as the feature distance L. The stability coefficient Q of the target person is calculated using the feature distance L and the circumcircle radius R, where Q = L / R. When the stability coefficient Q is less than the set stability coefficient threshold, the target person's behavior is determined to be abnormal.

5. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 1, characterized in that, The monitoring server determines the spatial feature parameters of the target person's behavior based on several posture key points. When determining whether the target person's behavior is abnormal based on the spatial feature parameters, it selects three target key points from all posture key points and uses these three target key points to construct the target joint angles of the target person. The target joint angles of the target person on the monitoring images corresponding to three consecutive time points are then substituted into a preset joint angular momentum feature calculation model. The calculation is performed to obtain the joint angular momentum characteristic value M of the target person. When the joint angular momentum characteristic value M of the target person exceeds the set angular momentum threshold, the target person's behavior is determined to be abnormal. , and These are the target joint angles of the target personnel in the personnel monitoring images at three consecutive time points, where Δt is the time interval between each time point, and I is the set equivalent inertia coefficient.

6. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 1, characterized in that, The monitoring server is also used to mark the warning area in the corresponding personnel monitoring image according to the preset warning area image coordinate parameters, and to determine whether the key points of the target personnel's posture are within the marked warning area. When it is determined that the key points of the target personnel's posture are within the marked warning area, the server generates area intrusion warning information and transmits the area intrusion warning information to the data communication device. The data communication device is also used to transmit the area intrusion warning information issued by the monitoring server to the alarm terminal. The alarm terminal is also used to receive the area intrusion warning information transmitted by the data communication device and to play and display the area intrusion warning information.

7. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 1, characterized in that, The monitoring server is also used to detect whether a target person in the personnel monitoring image is wearing a safety helmet, and to send a safety warning to the data communication device when it detects that the target person in the personnel monitoring image is not wearing a safety helmet.

8. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 7, characterized in that, When the monitoring server performs helmet detection on a target person in a personnel monitoring image, it extracts the target person image region from the personnel monitoring image, performs image enhancement and grayscale processing on the target person image region to obtain a grayscale image of the target person, and uses a pre-trained YOLOv11 model to perform helmet detection on the grayscale image of the target person to obtain the corresponding helmet wearing detection result.

9. The AI-based abnormal behavior recognition and early warning system for construction site safety supervision according to claim 1, characterized in that, The alarm terminal includes a voice alarm and a light alarm. The voice alarm is used to issue a warning voice based on the safety warning information, and the light alarm is used to issue a warning light based on the safety warning information.

Images