CN-121999341-A - Construction personnel unsafe behavior identification and early warning method and system based on improvement YOLOv11

CN121999341ACN 121999341 ACN121999341 ACN 121999341ACN-121999341-A

Abstract

The invention provides an improved YOLOv 11-based construction personnel unsafe behavior identification and early warning method and system, which belong to the technical field of building construction safety risk identification, acquire construction scene video data to be processed, and process the acquired construction scene video data to be processed by utilizing a pre-trained identification model to obtain a construction personnel unsafe behavior identification result. The invention solves the problems of dependence on manpower, poor real-time performance, low accuracy and weak small target detection capability of the traditional construction safety monitoring, provides high-efficiency and intelligent technical support for construction safety risk early warning, and opens up a new path for construction safety management.

Inventors

SHI XIAOBO
WANG YANG
WANG JIANMING
LI XUDONG
LI XIAOYUN
WANG JIANQIAO
Gao Fanqian
LI CHONG
SUN ZUO
PU ZHIZHOU
WANG YANHUI
MA JIANGPING
DU XIAOWEI
WANG DAWEI
SUN BOJIE
SU WEI
Bu Weicheng
WANG CHUNJIAO

Assignees

张家口通泰大数据信息服务有限公司
北京交通大学

Dates

Publication Date: 20260508
Application Date: 20260126

Claims (10)

1. The method for identifying and early warning unsafe behaviors of constructors based on improvement YOLOv is characterized by comprising the following steps: acquiring construction scene video data to be processed; The method comprises the steps of processing acquired construction scene video data to be processed by utilizing a pre-trained identification model to obtain an identification result of unsafe behaviors of constructors, wherein the training of the identification model comprises the steps of preprocessing construction scene video images, optimizing image quality in a complex environment, constructing an unsafe behavior data set and a characteristic set of the constructors in different construction scenes, defining identification objects and judging rules, extracting constructors and safety related target characteristics based on an improved YOLOv algorithm, integrating YOLOv11 target detection and ByteTrack multi-target tracking technology, extracting constructor track characteristics, identifying static offensiveness and dynamic risk events, integrating static target characteristics, gesture characteristics and track characteristics, stabilizing alarms through a time smoothing and hysteresis mechanism, and outputting unsafe behavior identification results and classification information.
2. The method for identifying and early warning the unsafe behavior of the constructor based on the improvement YOLOv11, which is characterized in that an unsafe behavior data set under different scenes is constructed, and identification objects and judgment rules are defined, wherein the method comprises the steps of combining typical risks and data realizabilities of construction sites, dividing unsafe behavior into two main categories, namely static violations and dynamic risk events, wherein the static violations refer to violation states or behaviors which can be judged through visual clues at given moments, and selecting four types of static violations, namely not wearing a safety helmet, not wearing a reflective garment/a safety vest, smoking and playing a mobile phone; let t be the input video frame at time t, the detector outputs a set of detection results: ; Wherein the method comprises the steps of A bounding box is represented and a bounding box is represented, The category is indicated as such, The confidence score is represented as a function of the confidence score, Representing the number of detection targets; Static violations are determined at the personnel level, for each personnel bounding box Safety-related objects are related by spatial association rules Assigned to the person, the association rule includes IoU Either the object center point is located within the personnel box or the object box is contained within the personnel box.
3. The method for identifying and pre-warning unsafe behaviors of constructors based on the improvement YOLOv, according to claim 1, wherein the method for achieving static target characteristics of constructors based on the improvement YOLOv model comprises the steps of adopting a single-stage backbox-Neck-Head design architecture, wherein the backbox is responsible for extracting multi-scale characteristics, neck enhancing multi-scale representation through characteristic fusion, and the Head outputs classification and bounding box regression results on multiple scales, To reduce redundant computation in a Backbone, a lightweight module named C3k2-Faster is proposed by integrating partial convolutions given input features The channels are divided into Multiple convolution channels An identity mapping channel, wherein The output formula is: ; Wherein the method comprises the steps of Representing a cascade of channels, an optional 1 x1 convolution is used for feature fusion/alignment.
4. The method for identifying and pre-warning unsafe behaviors of constructors based on the improvement YOLOv as claimed in claim 1, wherein the step of fusing YOLO target detection and ByteTrack tracking technology to realize behavior identification comprises the steps of: The method comprises the steps of carrying out multi-target tracking under a Kalman prediction and Hungary matching framework by ByteTrack, further improving track continuity under a shielding or crowding scene by utilizing a low confidence detection result, predicting a target track through Kalman filtering, carrying out two-stage association matching, solving the problem of target loss caused by shielding, and generating a continuous and stable constructor motion track: Dynamic events rely on time-continuous trajectory information, the trajectory of worker k being expressed as: ; Wherein the method comprises the steps of Is the boundary box tracked at the moment t and uses the reference point To formulate event constraints; and selecting an Adam optimizer as a strategy for adjusting model parameters, and simultaneously selecting cross entropy as a loss function to judge the performance of the model on a sample.
5. The method for identifying and pre-warning unsafe behaviors of constructors based on the improvement YOLOv as claimed in claim 1, wherein the method for merging static target features, gesture features and track features, stabilizing an alarm through a time smoothing and hysteresis mechanism and outputting unsafe behavior identification results and classification information comprises the following steps: The early warning mechanism for two types of dynamic risk events is as follows: A mechanical proximity risk, which characterizes unsafe exposure conditions when the worker is too close to the machine/vehicle, for worker k and machine/vehicle m, set their reference points at time t as And The distance is defined as: ; if the distance is Duration of at least Triggering a proximity event alert if a frame, if a distance Duration of at least And the frame is used for removing the alarm, and the threshold value is configured with scale perception normalization in order to keep robustness under different visual angles and scales.
6. The method for identifying and pre-warning unsafe behavior of constructor based on the improvement YOLOv as set forth in claim 5, wherein the forbidden zone is invaded by setting the dangerous/limiting area as a polygon If the reference point Duration of at least Triggering an intrusion event alert if a frame Duration of at least A frame, then the alarm is released; Time smoothing and hysteresis policy detection jitter, occlusion, and boundary wander for alarm stabilization may cause frequent switching of risk states, causing alarm flickering, first applying time smoothing to transient risk decisions, followed by applying hysteresis update rules to determine alarm states : ; Wherein, the And The minimum duration of triggering and releasing the alarm, respectively, suppresses flicker caused by short-term fluctuations by introducing a temporal "forbidden zone".
7. An improved YOLOv-based constructor unsafe behavior identification and early warning system, comprising: The acquisition module is used for acquiring the construction scene video data to be processed; The processing module is used for processing the acquired construction scene video data to be processed by utilizing a pre-trained identification model to obtain an identification result of unsafe behaviors of constructors, wherein the training of the identification model comprises the steps of preprocessing construction scene video images, optimizing image quality in complex environments, constructing an unsafe behavior data set and a characteristic set of the constructors in different construction scenes, defining identification objects and judging rules, extracting constructors and safety related target characteristics based on an improved YOLOv algorithm, fusing an improved YOLOv target detection and ByteTrack multi-target tracking technology, extracting track characteristics of the constructors, identifying static violations and dynamic risk events, fusing the static target characteristics, gesture characteristics and track characteristics, stabilizing an alarm through a time smoothing and hysteresis mechanism, and outputting the identification result of unsafe behaviors and classification information.
8. A non-transitory computer readable storage medium storing computer instructions which, when executed by a processor, implement the constructor unsafe behavior recognition and early warning method based on the improvement YOLOv11 of any one of claims 1-6.
9. A computer device comprising a memory and a processor, the processor and the memory in communication with each other, the memory storing program instructions executable by the processor, the processor invoking the program instructions to perform the constructor unsafe behavior recognition and early warning method based on the improvement YOLOv according to any one of claims 1-6.
10. An electronic device comprising a processor, a memory and a computer program, wherein the processor is coupled to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the constructor unsafe behavior recognition and early warning method based on the improvement YOLOv according to any one of claims 1-6.

Description

Construction personnel unsafe behavior identification and early warning method and system based on improvement YOLOv11 Technical Field The invention relates to the technical field of building construction safety risk identification, in particular to a constructor unsafe behavior identification and early warning method and system based on improvement YOLOv 11. Background The construction scene of building construction and traffic infrastructure has the characteristics of complex operation environment, frequent cross operation, high personnel mobility and dense risk points, and unsafe behaviors of constructors (such as no wearing of safety measures, smoking, using mobile phones, closely contacting with machinery, rushing into a limited area and the like) are main causes of accidents. Through investigation, construction safety monitoring mainly depends on manual inspection and traditional video playback inspection, and has the obvious defects of poor real-time performance, difficulty in covering a full scene during manual inspection, incapability of timely early warning after an accident occurs, low accuracy, easiness in fatigue and subjective factor influence due to experience judgment of workers, high omission rate and false detection rate, weak recognition capability on small targets (cigarettes and mobile phones), poor pertinence, suitability for general scenes, weak recognition capability on special unsafe behaviors (such as short-distance contact machinery and intrusion limit areas) of different construction scenes, poor alarm stability, easiness in detection jitter and boundary loitering influence due to alarm flickering due to dynamic event judgment, low efficiency, large-scale construction sites needing to input a large amount of manpower, and high monitoring cost. The existing computer vision technology has achieved a certain achievement in the fields of traffic, security and protection, but the construction scene has the special problems of changeable illumination, serious shielding, complex behavior types, dense small targets and the like, and lacks a special data set and multi-algorithm fusion scheme aiming at four kinds of core risk scenes, so that the existing technology is difficult to directly adapt. Therefore, there is a need for a method for identifying unsafe behaviors of constructors aiming at construction core risk scenes, integrating advantages of multiple algorithms, being accurate and efficient, and having stable alarms, and solving the pain points of traditional monitoring. Disclosure of Invention The invention aims to provide a method and a system for identifying and early warning unsafe behaviors of constructors based on improvement YOLOv, which are used for solving at least one technical problem in the background technology. In order to achieve the above purpose, the present invention adopts the following technical scheme: in a first aspect, the present invention provides a method for identifying and early warning unsafe behaviors of constructors based on improvement YOLOv, comprising: acquiring construction scene video data to be processed; The method comprises the steps of processing acquired construction scene video data to be processed by utilizing a pre-trained identification model to obtain an identification result of unsafe behaviors of constructors, wherein the training of the identification model comprises the steps of preprocessing construction scene video images, optimizing image quality in a complex environment, constructing an unsafe behavior data set and a characteristic set of the constructors in different construction scenes, defining identification objects and judging rules, extracting constructors and safety related target characteristics based on an improved YOLOv algorithm, integrating YOLOv11 target detection and ByteTrack multi-target tracking technology, extracting constructor track characteristics, identifying static offensiveness and dynamic risk events, integrating static target characteristics, gesture characteristics and track characteristics, stabilizing alarms through a time smoothing and hysteresis mechanism, and outputting unsafe behavior identification results and classification information. As a further limitation of the first aspect of the invention, constructing unsafe behavior data sets under different scenes, defining identification objects and judging rules, wherein the definition of identification objects and judging rules comprises that combining typical risks and data realizabilities of construction sites, classifying unsafe behaviors into two main categories, namely static violations and dynamic risk events, wherein the static violations refer to violation states or behaviors which can be judged through visual clues at given moments, and selecting four types of static violations, namely, not wearing safety helmets, not wearing reflective clothing/safety vests, smoking and playing mobile phones; let t be the input video frame at time t, the