A target recognition-based collapse risk early warning method

Abstract

The application discloses a target recognition-based collapse risk early warning method, which comprises the following steps: S1, delimiting a monitoring area, calibrating a reference object in the monitoring area, and continuously acquiring image data with time sequence characteristics of the monitoring area through a monitoring device; S2, continuously monitoring and calculating displacement and deformation of the reference object based on the image data acquired in S1, and determining that the reference object is abnormal when the displacement and deformation of the reference object exceed a preset value; S3, acquiring weather, construction and maintenance information in the monitoring area, and filtering the abnormal determination in S2 according to the information; S4, when it is still determined that the reference object is abnormal after the filtering in S3, classifying displacement and deformation characteristics of abnormal image data, and determining a disaster category based on the classification of the displacement and deformation characteristics; and S5, determining and finally outputting a disaster type, a precise disaster position, a disaster influence range and a disaster warning grade based on the disaster category in S4 and the displacement and deformation of the reference object in S2.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, specifically to a method for early warning of collapse risk based on target recognition. Background Technology

[0002] Slope collapse and bridge collapse are two highly destructive engineering geological hazards, causing not only enormous property losses but also directly threatening the lives of the people. Therefore, intelligent identification and monitoring of these two types of disasters is of irreplaceable necessity. With the rapid advancement of infrastructure construction in my country, numerous bridges and slopes are distributed across key areas such as highways, railways, and water conservancy projects. Some of these facilities are subjected to complex geological environments and frequent loads for extended periods, leading to increasingly prominent aging and damage issues. Traditional manual monitoring methods are no longer sufficient to meet safety and prevention needs. Traditional monitoring relies on manual inspections and fixed-point detection, which suffers from drawbacks such as delayed response, limited coverage, and significant human error. Problems are often only discovered after a disaster occurs, failing to provide early warnings and proactive prevention. In contrast, intelligent identification technology, relying on technologies such as the Internet of Things, artificial intelligence, and big data, can achieve 24 / 7 uninterrupted monitoring, capturing subtle anomalies such as slope displacement, crack propagation, and bridge structural deformation in real time. This allows for accurate identification of potential disaster hazards, significantly improving monitoring efficiency and accuracy, and buying valuable time for disaster prevention and control. From the perspective of people's livelihood, slope collapses and bridge collapses often occur along major transportation routes and around towns. Once they happen, they can lead to traffic disruptions, casualties, and severely impact the normal production and lives of the people, even triggering secondary disasters. Intelligent identification can predict disaster risks in advance, issue timely warnings, guide the evacuation of people, and control traffic, minimizing casualties and property losses and protecting the safety of people's lives and property. From the perspective of engineering operation and maintenance and socio-economic development, intelligent identification can enable early detection and management of hidden dangers, reduce facility maintenance costs, extend the service life of projects, and avoid huge economic losses caused by infrastructure paralysis due to disasters. At the same time, its precise monitoring capabilities can provide scientific data support for engineering design and maintenance plan optimization, promote the development of infrastructure construction towards safety, intelligence, and precision, help build a solid safety production line, and ensure the sustained and stable development of the social economy. Summary of the Invention

[0003] The purpose of this application is to provide a collapse risk early warning method based on target recognition, and the specific technical solution is as follows:

[0004] A collapse risk early warning method based on target recognition includes: S1, delineating a monitoring area and marking reference objects within the monitoring area, and continuously acquiring time-series image data of the monitoring area through monitoring equipment; S2, based on the image data acquired in S1, continuously monitoring and calculating the displacement and deformation of the reference objects, and determining that the reference objects are abnormal when the displacement and deformation exceed preset values; S3, acquiring weather, construction, and maintenance information within the monitoring area, and filtering the abnormality determination in S2 based on this information; S4, if the reference objects are still determined to be abnormal after filtering in S3, classifying the abnormal image data by displacement and deformation features, and determining the disaster category based on the classification of displacement and deformation features; S5, based on the disaster category in S4 and the displacement and deformation of the reference objects in S2, determining and finally outputting the disaster type, precise disaster location, disaster impact range, and disaster early warning level.

[0005] When delineating monitoring areas and setting reference objects in S1, the following are included: S1.1 For slope areas, the exposed rock mass, slope protection, and slope toe guardrail are designated as monitoring areas, and the exposed rock mass, artificial targets, slope protection masonry, and / or slope toe guardrail are marked as reference objects; S1.2 For bridge areas, the bridge bearings, bridge deck guardrails, and bridge main beams are designated as monitoring areas, and the bridge bearings, bridge deck expansion joints, and / or guardrail posts are marked as reference objects.

[0006] S2 includes the following steps for monitoring and calculating the displacement and deformation of the reference object: S2.1. Using a YOLOv8 model deployed based on the Darknet framework, the model interface is called via C++ to accurately detect the reference object within the monitoring area in each frame, lock the reference object's position, and achieve continuous tracking of the reference object through an inter-frame coordinate association algorithm, avoiding tracking loss due to changes in lighting and slight vibrations; S2.2. Using the initial coordinates of the reference object as a reference, coordinate calculation logic is written in C++ to calculate the coordinate offset of the reference object in the image in real time, including lateral and longitudinal displacement, and obtain and record the displacement change time series data; S2.3. Using the initial shape of the reference object as a reference, the changes in the shape of the reference object in the image are monitored in real time. Combined with the displacement change time series data in S2.2, it is determined whether the reference object has undergone tensile, compressive, or fracture structural deformation, and the structural deformation is calculated; S2.4. Based on the classification of the monitoring area and the classification of the reference object in S1, anomaly thresholds for reference object displacement and deformation are set. When the displacement in S2.2 and the deformation in S2.3 exceed the anomaly threshold, the reference object is determined to be abnormal.

[0007] When filtering anomaly determinations in S3, the following steps are taken: S3.1, based on the anomaly determination in S2.4 and publicly available weather information, exclude non-structural deformations of reference objects caused by natural factors such as heavy rain and strong winds, and retain continuous and trend-based displacement and deformation anomalies; S3.2, based on S3.1, exclude reference object movement caused by construction and maintenance operations based on construction and maintenance information; S3.3, based on S3.2, exclude reference object anomalies caused by occlusion and identification errors.

[0008] When determining the disaster category in S4, the following criteria are included: S4.1 When the reference object slides downward or shifts towards the road surface, and the deformation manifests as rock fracture or slope fall-off, and the displacement / deformation continues to expand and exceeds the set abnormal threshold, it is determined to be a slope collapse; S4.2 When the reference object sinks vertically or shifts, and the deformation manifests as main beam bending, support fracture, or guardrail fall-off, and the displacement / deformation continues to expand and exceeds the set abnormal threshold, it is determined to be a bridge collapse.

[0009] The beneficial effect of this application lies in its ability to accurately identify two types of geological disasters—slope collapse and bridge collapse—by continuously monitoring the displacement and deformation of reference objects in slope and bridge areas. This is applicable to real-time monitoring of highway slopes and bridge sections. Before and during a slope collapse or bridge collapse, reference objects such as exposed rock masses, targets, and artificial structures (guardrails, slope protection, bridge bearings, etc.) in the slope area will exhibit significant displacement and deformation. Through video analysis technology, using a YOLO model to detect and lock onto these reference objects, and continuously tracking their displacement and deformation characteristics, after eliminating interference factors, the occurrence and severity of slope collapse and bridge collapse events can be accurately determined. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the application process. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to specific embodiments and accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of this application. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0012] A collapse risk early warning method based on target recognition includes:

[0013] S1. Delineate the monitoring area and mark reference objects within the monitoring area, and continuously acquire time-series image data of the monitoring area through monitoring equipment. Specifically, delineating the monitoring area and marking reference objects includes: S1.1. For slope areas, delineate the exposed rock mass, slope protection, and slope toe guardrail as the monitoring area, and mark the exposed rock mass, artificial targets, slope protection masonry, and / or slope toe guardrail as reference objects; S1.2. For bridge areas, delineate the bridge bearings, bridge deck guardrails, and bridge main beams as the monitoring area, and mark the bridge bearings, bridge deck expansion joints, and / or guardrail posts as reference objects.

[0014] S2. Based on the image data obtained in S1, continuously monitor and calculate the displacement and deformation of the reference object. When the displacement and deformation of the reference object exceed the preset value, the reference object is determined to be abnormal. Specifically, the monitoring and calculation of the displacement and deformation of the reference object includes: S2.1, using a YOLOv8 model deployed based on the Darknet framework, calling the model interface through C++, accurately detecting the reference object within the monitoring area in each frame, locking the reference object's position, and achieving continuous tracking of the reference object through an inter-frame coordinate association algorithm to avoid tracking loss due to changes in lighting and slight vibrations; S2.2, using the initial coordinates of the reference object as a reference, writing coordinate calculation logic in C++ to calculate the coordinate offset of the reference object in the image in real time, including lateral and longitudinal displacement, obtaining and recording the displacement change time series data; S2.3, using the initial shape of the reference object as a reference, monitoring the changes in the shape of the reference object in the image in real time, combining the displacement change time series data in S2.2, determining whether the reference object has undergone tensile, compressive, or fracture structural deformation, and calculating the structural deformation; S2.4, setting abnormal thresholds for reference object displacement and deformation based on the monitoring area classification and reference object calibration classification in S1, determining that the reference object is abnormal when the displacement in S2.2 and the deformation in S2.3 exceed the abnormal threshold.

[0015] S3. Obtain weather, construction, and maintenance information within the monitoring area, and filter the anomaly determinations in S2 based on this information. Specifically, the anomaly determination filtering includes: S3.1. Based on the anomaly determinations in S2.4 and publicly available weather information, exclude non-structural deformations of reference objects caused by natural factors such as heavy rain and strong winds, and retain continuous and trend-based displacement and deformation anomalies; S3.2. Based on S3.1, and based on construction and maintenance information, exclude reference object movement caused by construction and maintenance operations; S3.3. Based on S3.2, exclude reference object anomalies caused by obstruction and identification errors.

[0016] S4. If the reference object is still determined to be abnormal after filtering in S3, the abnormal image data is classified according to displacement and deformation features, and the disaster category is determined based on the classification of displacement and deformation features. Specifically, the disaster category determination includes: S4.1. When the reference object slides downward or shifts towards the road surface, and the deformation is manifested as rock fracture or slope collapse, and the displacement / deformation continues to expand, exceeding the set abnormal threshold, it is determined to be a slope collapse; S4.2. When the reference object sinks vertically or shifts, and the deformation is manifested as main beam bending, support fracture, or guardrail detachment, and the displacement / deformation continues to expand, exceeding the set abnormal threshold, it is determined to be a bridge collapse.

[0017] S5, based on the disaster category in S4 and the displacement and deformation of the reference objects in S2, determines and ultimately outputs the disaster type, precise location, impact range, and warning level. Specifically, it assesses confidence level, number of monitoring frames, reference object displacement / deformation data, accuracy of the YOLOv8 model based on the Darknet framework, and screenshots of typical anomalies (reference object deformation). All output results are displayed through an interactive interface built with the Qt framework, and data export is also supported. This provides accurate data for subsequent on-site investigation, emergency response, and traffic control. The overall functionality is implemented on the Ubuntu system, using C++ and the Qt framework, ensuring high operational efficiency.

Claims

1. A method for early warning of collapse risk based on target recognition, characterized in that, include: S1. Delineate the monitoring area and mark reference objects within the monitoring area, and continuously acquire image data with temporal characteristics of the monitoring area through monitoring equipment; S2. Based on the image data obtained in S1, continuously monitor and calculate the displacement and deformation of the reference object. When the displacement and deformation of the reference object exceed the preset value, determine that the reference object is abnormal. S3. Obtain weather, construction and maintenance information within the monitoring area, and filter the anomaly determination in S2 based on this information; S4. If the reference object is still determined to be abnormal after filtering in S3, the abnormal image data is classified according to displacement and deformation features, and the disaster category is determined based on the classification of displacement and deformation features. S5. Based on the disaster category in S4 and the displacement and deformation of the reference object in S2, determine and finally output the disaster type, precise location of the disaster, scope of the disaster impact, and disaster warning level.

2. The collapse risk early warning method based on target recognition as described in claim 1, characterized in that, When defining the monitoring area and calibration reference in S1, the following is included: S1.1 For the slope area, the exposed rock mass, slope protection and slope toe guardrail of the slope are designated as the monitoring area, and the exposed rock mass, artificial target, slope protection masonry and / or slope toe guardrail are marked as reference objects. S1.2 For the bridge area, the bridge bearings, bridge deck railings, and bridge main beams are designated as monitoring areas, and the bridge bearings, bridge deck expansion joints, and / or railing posts are marked as reference objects.

3. The collapse risk early warning method based on target recognition as described in claim 2, characterized in that, The monitoring and calculation of the displacement and deformation of the reference object in S2 includes: S2.

1. The YOLOv8 model based on the Darknet framework is adopted. The model interface is called through C++. The reference objects in the monitoring area are accurately detected in each frame, the position of the reference objects is locked, and the reference objects are continuously tracked through the inter-frame coordinate association algorithm to avoid tracking loss due to changes in lighting and slight vibrations. S2.

2. Using the initial coordinates of the reference object as a reference, coordinate calculation logic is written in C++ to calculate the coordinate offset of the reference object in the image in real time, including horizontal and vertical displacement, and to obtain and record the displacement change time series data. S2.

3. Based on the initial shape of the reference object, monitor the changes in the shape of the reference object in the image in real time. Combined with the displacement change time series data in S2.2, determine whether the reference object has undergone tensile, compressive, or fracture structural deformation and calculate the structural deformation. S2.

4. Based on the classification of the monitoring area and the classification of the reference object in S1, set the threshold for abnormal displacement and deformation of the reference object. When the displacement in S2.2 and the deformation in S2.3 exceed the threshold, the reference object is determined to be abnormal.

4. The collapse risk early warning method based on target recognition as described in claim 3, characterized in that, The filtering of anomaly determination in S3 includes: S3.1 Based on the anomaly determination in S2.4 and the publicly available weather information, non-structural deformation of the reference object caused by natural factors such as heavy rain and strong winds is excluded, while continuous and trend-based displacement and deformation anomalies are retained. S3.2 Based on S3.1, and based on construction and maintenance information, exclude the movement of reference objects caused by construction and maintenance operations; S3.

3. Based on S3.2, eliminate reference object anomalies caused by occlusion and identification errors.

5. The collapse risk early warning method based on target recognition as described in claim 4, characterized in that, The determination of the disaster category in S4 includes: S4.1 When the reference object slides downward or shifts towards the road surface, and the deformation manifests as rock fracture and slope fall-off, and the displacement / deformation continues to expand and exceeds the set abnormal threshold, it is judged as a slope collapse. S4.2 When the reference object shows vertical sinking or displacement, and the deformation manifests as bending of the main beam, breakage of the support, and detachment of the guardrail, and the displacement / deformation continues to expand and exceeds the set abnormal threshold, it is judged as a bridge collapse.

Images