Method for collapse and landslide dumping alarm based on monitoring video stream and distance-direction angle relationship

Abstract

The present application belongs to the field of geological disaster and engineering safety monitoring technology, and discloses a kind of based on monitoring video stream and distance-direction angle relationship's collapse and slide and dump alarm method, and collapse and slide and dump alarm method uses a kind of collapse and slide and dump alarm system, including monitoring stream acquisition device, video stream processing module, distance-direction angle relationship calculation module and alarm output module;Specifically including the following steps: S1, acquisition monitoring real-time video stream;S2, calibrate camera parameters, and select reference point, each monitoring object key monitoring point;S3, dynamic tracking monitoring point and analyzing sliding situation;S4, according to monitoring point sliding situation analysis each monitoring object and overall sliding situation;S5, determine landslide early warning level and carry out sound and light alarm.The present application does not need to deploy expensive sensor, and is low in cost, fast in response, strong in adaptability, can greatly improve the timeliness and reliability of disaster early warning.

Description

Technical Field

[0001] This invention belongs to the field of geological disaster and engineering safety monitoring technology, specifically relating to a landslide and collapse alarm method based on the relationship between monitoring video stream and distance-direction angle. Background Technology

[0002] Landslides and collapses encompass geological and engineering disasters such as rock and soil collapses, natural and engineering slope slippage, and the toppling of high-altitude unstable rock masses and engineering components. Landslides and collapses are a global geological and engineering hazard, frequently occurring on slopes and cut slopes in mountainous and hilly areas, as well as in sensitive areas such as engineering embankments, building foundations, and reservoir banks. They cause enormous losses to industrial and agricultural production, as well as to people's lives and property, sometimes even leading to devastating catastrophes.

[0003] Since the mid-20th century, humans have been researching monitoring technologies for landslides and unstable rockfalls. With continuous technological advancements, monitoring methods have evolved, including Global Navigation Satellite System (GNSS) and fiber optic cable displacement gauges. Currently, GNSS monitoring technology is one of the most widely used methods for monitoring landslide and collapse deformation. However, due to the inherent characteristics of landslides and collapses, these methods all have certain limitations. Specifically:

[0004] Landslides and collapses generally do not occur suddenly; their development and eventual failure involve a process of initial deformation, uniform sliding, accelerated sliding, and instability failure. The initial deformation and uniform sliding phases are typically lengthy and less dangerous. The accelerated sliding and instability failure phases, however, develop rapidly and are extremely dangerous, representing the critical period for emergency response and personnel evacuation. GNSS typically takes 3-10 minutes to process acquired data, indicating that the data obtained at any given moment reflects deformation results from 3-10 minutes prior to the landslide. Therefore, GNSS data is often delayed when a sudden change occurs, potentially missing the critical warning window. Furthermore, to achieve millimeter-level monitoring accuracy, GNSS often requires differential configurations or fixed base stations, which not only increases the overall system cost but also limits deployment in complex terrain, poor communication, and spatially constrained mountainous and wilderness environments. For example, forest obstruction, building interference, or terrain occlusion can all affect GNSS signal quality, reducing system stability and reliability.

[0005] Therefore, traditional monitoring methods based on physical contact sensors suffer from problems such as low monitoring frequency, high deployment cost, and poor environmental adaptability when facing the identification of early signs of rapid landslides and collapses and real-time alarms. There is an urgent need for a new type of landslide and collapse monitoring and alarm technology that is non-contact, has a rapid response, is flexible in deployment, and is cost-controllable. Summary of the Invention

[0006] The purpose of this invention is to provide a landslide and collapse alarm method based on the relationship between monitoring video stream and distance-direction angle, so as to solve the problems of high monitoring cost and untimely monitoring and early warning in the prior art mentioned in the background.

[0007] To achieve the above objectives, this invention provides a landslide and collapse alarm method based on the relationship between monitoring video stream and distance-direction angle. It uses a landslide and collapse alarm system to perform real-time monitoring and landslide and collapse alarm of potential landslide and collapse areas. The landslide and collapse alarm system includes a monitoring stream acquisition device, a video stream processing module, a distance-direction angle calculation module, and an alarm output module, which are used to realize a closed-loop process of monitoring video stream acquisition, processing, anomaly detection, and alarm output, respectively.

[0008] The landslide and collapse alarm method includes the following steps:

[0009] S1. Acquire real-time video streams from monitoring systems;

[0010] S2. Calibrate the camera parameters and select reference points and key monitoring points for each monitored object;

[0011] S3. Dynamically track and monitor points and analyze the slippage situation;

[0012] S4. Analyze the slippage trend of each monitored object and the overall slippage trend based on the slippage trend of the monitoring points;

[0013] S5. Determine the landslide warning level and issue an audible and visual alarm.

[0014] In one specific implementation, in step S1, a monitoring stream acquisition device is deployed in the safe and stable zone opposite the potential landslide and collapse zone to acquire monitoring video streams in real time and transmit them to the video stream processing module; the device used to acquire the video streams is a monitoring camera, which includes a high-definition CCD monitoring camera and an infrared camera.

[0015] In one specific implementation, in step S2, the user needs to input basic parameters of the monitoring area. Based on these basic parameters, the landslide and collapse alarm system uses the principle of similar triangles to construct a mapping relationship between the actual terrain displacement and the pixel offset in the image. This mapping relationship is used to compare the pixel-level changes of key points relative to reference points in subsequent images, and to determine whether there are signs of landslides or collapses at the monitoring points.

[0016] In one specific implementation, in step S2, the reference point is a relatively stable and unchanging area in the image, including exposed rock in the stable area, pile foundations, and fixed structures; the monitoring point is a key feature point in an area with significant risk of landslides and collapses, including rock cracks, protruding points of hidden danger bodies, and block edges.

[0017] In one specific implementation, in step S3, the monitoring point tracking algorithm used when dynamically tracking the monitoring point includes a dynamic coupling algorithm that combines Kalman filtering and optical flow: the position of the monitoring point in the current frame is predicted according to Kalman filtering, and the predicted position is used as the initial search point of optical flow; and the coordinates of the monitoring point are further corrected using template matching method.

[0018] In one specific implementation, step S3 involves calculating the spatial relationship change of each monitoring point relative to the reference point in real time, specifically including:

[0019] In the feature point tracking process, the pixel coordinates (x, y, y) of the monitoring point are extracted frame by frame from the video frame. t ,y t ) and the corresponding reference point pixel coordinates (x r0 ,y r0 ), calculates the changes in distance and orientation angle between the current image monitoring point and the initial image reference point in real time;

[0020] The pixel distance change Δd is expressed as:

[0021] ;

[0022] The pixel orientation angle change Δθ is expressed as:

[0023] ;

[0024] Where (x) t ,y t (x) represents the pixel coordinates of the monitoring point at time t in the current frame. r0 ,y r0 () represents the pixel coordinates of the reference point in the initial frame;

[0025] The above distance change Δd is compared with the calibrated pixel displacement threshold T. d，pix The change in orientation angle Δθ and the calibrated pixel orientation angle threshold T θ，pix The comparison is performed; if any item exceeds the corresponding threshold, it is considered that there is a sign of landslide or collapse at that monitoring point.

[0026] In one specific implementation, in step S3, when a monitoring object has multiple reference points, the spatial relationship between the monitoring point of the monitoring object and its corresponding set of reference points is evaluated. The system calculates the distance change Δd and the direction angle change Δθ between the monitoring point and each reference point respectively. The above multiple sets of distance changes and direction angle changes are fused, and the fusion method includes the maximum value and the average value.

[0027] In step S3, the signs of collapse and toppling at the monitoring point include the maximum displacement, average displacement, maximum change in orientation angle, and average change in orientation angle.

[0028] In one specific implementation, in step S4, the factors for determining the risk level of a single monitored object include the volume of landslide and collapse at the monitoring point, the maximum displacement at the monitoring point, the average displacement, the maximum change in direction angle, and the average change in direction angle; the factors for determining the overall landslide and collapse risk, i.e., the alarm level, include the weighted risk level, the average risk level, and the maximum risk level of the monitored object.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] This invention eliminates the need for expensive sensors, is low-cost, fast-responding, and highly adaptable, and can significantly improve the timeliness and reliability of disaster early warning.

[0031] The landslide and collapse alarm method proposed in this invention is based on video stream monitoring. It can be deployed using conventional video surveillance equipment, without relying on dedicated sensors, making it convenient, low-cost, and highly adaptable. Through video stream analysis, it can achieve long-distance, non-contact identification of landslide and collapse deformation, making it particularly suitable for complex terrain areas such as steep slopes, surrounding rock masses, and landslide zones where on-site sensor deployment is unsuitable. The system identifies signs of landslides and collapses by analyzing changes in the spatial relationship (distance and azimuth angle) between reference points and monitoring points in the image, significantly improving the stability and reliability of monitoring.

[0032] 1. This invention constructs a hierarchical model of "hazard body - reference point - monitoring point - weight" and introduces weighting factors to take into account the relative importance of multiple monitoring objects in the overall alarm judgment. The system supports parallel tracking of multiple monitoring objects, combines their respective collapse and toppling trends for comprehensive judgment, and then outputs the overall collapse and toppling risk assessment level of the monitoring area. This achieves the ability to perceive the evolution from local features to regional risks, effectively avoiding the problem of false alarms or missed alarms for single-point anomalies, and improving the accuracy and robustness of the overall alarm.

[0033] 2. To address the issues of low search efficiency and susceptibility to interference in traditional optical flow methods under complex terrain or varying lighting conditions, the system introduces a Kalman prediction mechanism. This mechanism predicts the possible locations of monitoring points in each image frame, performing optical flow searches only within a small area, thus significantly improving the algorithm's real-time performance and stability. Simultaneously, the optical flow results influence the Kalman state update, forming a closed-loop feedback loop that makes the prediction more adaptive.

[0034] 3. Supports outputting alarm results via audio-visual equipment, allowing on-site personnel to prepare in advance and evacuate in an emergency. It also records information such as the displacement, direction angle, and alarm level of the monitoring point, facilitating the analysis of early signs of landslides and collapses in the later stages.

[0035] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The present invention will now be described in further detail. Attached Figure Description

[0036] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0037] Figure 1 This is a flowchart illustrating one embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of monitoring point selection according to an embodiment of the present invention; in the figure, R1, R2, R3, R4, R5, and R6 are different reference points on the exposed rock in the stable area, P1-1, P1-2, and P1-3 are different monitoring points of monitoring object 1, which is the protruding point of the hidden danger body, and its weight is wi1; P2-1, P2-2, P2-3, and P2-4 are different monitoring points of monitoring object 2, which is the edge of the block, and its weight is wi2;

[0039] Figure 3 This is a schematic diagram of the monitoring point tracking process according to an embodiment of the present invention. Detailed Implementation

[0040] The embodiments of the present invention will be described in detail below. The specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0041] Example 1

[0042] like Figure 1 As shown, an embodiment of the present invention provides a method for landslide and collapse alarm based on the relationship between monitoring video stream and distance-direction angle, the steps of which include:

[0043] S1. Deploy a monitoring stream acquisition device in a safe and stable area opposite the potential landslide and collapse area to acquire monitoring video streams in real time and transmit them to the video stream processing module.

[0044] S2. The user inputs the basic parameters of the monitoring area to complete the system calibration. Initial images are extracted from the video stream. The user first selects prominent feature points of typical stable and unchanging features in the image (such as exposed bedrock corners, pile foundation apexes, etc.) as reference points. Then, the system groups the monitoring objects according to potential hazards, selecting typical feature points of key parts of the monitoring objects, such as rock fissures, prominent hazard points, and block edge structures, as key monitoring points. The system assigns an independent color code to each monitoring object and associates its monitoring points with 2-3 reference points. The system sets the weight w for each monitoring object. iAnd perform normalization: Finally, a correlation model of target-weight-reference frame is constructed, such as... Figure 2 As shown.

[0045] S3. The image of the selected monitoring point is used as the initial frame and input into the video stream processing module. The system tracks the monitoring point frame by frame using a monitoring point tracking algorithm. Based on the spatial position of the monitoring point and the reference point in the initial frame, the system calculates in real time the distance change Δd and the orientation angle change Δθ between the current image monitoring point and the initial image reference point. If any of these values ​​exceeds the calibrated pixel displacement threshold T... d，pix Pixel orientation angle threshold T θ，pix When this occurs, it is considered that there are signs of impending landslide or collapse at the monitoring point;

[0046] Pixel distance change:

[0047] ;

[0048] Pixel orientation angle change ;

[0049] Where (x) t ,y t (x) represents the pixel coordinates of the monitoring point at time t in the current frame. r0 ,y r0 () represents the pixel coordinates of the reference point in the initial frame.

[0050] S4. Based on the number of monitoring points showing signs of landslides and collapses, the changes in distance and direction angle, and the weighting factors of the respective monitoring objects, assess the landslide and collapse situation of each monitoring object and the overall monitoring area; and determine its risk level accordingly, and enter the corresponding alarm stage.

[0051] S5. Based on the assessment results, the system triggers a level 1-9 audible and visual alarm mechanism, sending alarm information to on-site personnel and surrounding areas in real time via audible and visual alarms, prompting workers, passing vehicles, and personnel to quickly evacuate the danger zone. Simultaneously, the system automatically records changes in the status of monitoring points and alarm event logs for subsequent analysis and tracing.

[0052] This embodiment uses a landslide and collapse alarm system for real-time monitoring and alarming of potential landslide and collapse areas. The landslide and collapse alarm system includes a monitoring stream acquisition device, a video stream processing module, a distance-direction angle relationship calculation module, and an alarm output module, which are used to realize a closed-loop process of monitoring video stream acquisition, processing, anomaly detection, and alarm output.

[0053] Specifically, step S1 includes:

[0054] S1.1 Hardware Deployment. Deploy surveillance cameras in a safe and stable area opposite the landslide-prone zone to monitor potential landslides and collapses. Cameras should be fixedly installed in locations with stable supporting structures and unobstructed views, ensuring complete coverage of the potential sliding / collapse area. The video stream should be transmitted in real-time to the video stream processing module. The preferred acquisition devices are high-definition CCD surveillance cameras and infrared cameras, which collect and transmit surveillance video streams day and night in real-time, providing a stable data source for subsequent processing.

[0055] Specifically, step 2 includes:

[0056] S2.1 Parameter Configuration: The user inputs basic parameters of the monitoring area, including the actual width W, height H, average distance D between the camera and the monitoring surface, and the preset distance alarm threshold T. d (Unit: meters), and the direction angle alarm threshold T θ (Unit: degree).

[0057] S2.2, Physical-Pixel Mapping Model:

[0058] Based on the input basic parameters, the system uses the principle of similar triangles to construct a mapping relationship between actual terrain displacement and pixel offset in the image, specifically including:

[0059] Obtain the resolution parameter N of the image x N y (Total number of pixels in width direction, total number of pixels in height direction);

[0060] Calculate the actual length of each unit pixel in the image in the horizontal and vertical directions respectively: ;

[0061] The user-defined actual distance threshold T d Converted to image pixel displacement threshold, it is expressed as:

[0062] ; Set the actual orientation angle threshold T as set by the user θ Approximately considered as a pixel orientation angle threshold: T θ，pix =T θ ;

[0063] The mapping relationship will be used to compare the pixel-level changes of key points relative to the reference point in subsequent images, in order to determine whether there are signs of landslides or collapses at the monitoring points.

[0064] S2.3, Monitoring point selection:

[0065] S2.3.1 The user inputs the basic parameters of the monitoring area to complete the system calibration. The initial image is extracted from the video stream. The user first selects several obvious feature points on stable ground features as reference points to make them stable and unchanged in the time series. Rigid ground features such as exposed bedrock, pile foundations, and the corners of structures are often selected.

[0066] S2.3.2. Then group the monitored objects and select typical feature points such as rock mass cracks, obvious hidden danger bodies, and block edge structures as key monitoring points in their key risk areas;

[0067] S2.3.3 The system module assigns an independent color code to each monitored object and associates 2 to 3 reference points with its monitoring point;

[0068] S2.3.4. The system assigns weights based on the risk of each feature, and automatically normalizes the input weights (weights are related to alarm levels; the higher the target weight, the higher the priority of response when triggered). ;

[0069] Each monitoring point uses an independent color identifier to track multiple targets in parallel; risk priorities are assessed by setting different weights to achieve priority alarms.

[0070] Specifically, step S3 includes:

[0071] S3.1 After the user completes the selection of monitoring points, the current image is used as the initial frame; subsequently, using the monitoring points selected in the initial frame, the monitoring points are tracked in subsequent frames of the video stream, such as... Figure 3 As shown.

[0072] S3.2, Tracking of monitoring points:

[0073] S3.2.1 Kalman filtering predicts the position of the monitoring point in the current frame based on historical trajectories; Kalman filter state. The definition includes position (x, y) and velocity (v). x ,v y In a two-dimensional model, prediction and updating employ a state transition model, based on the state x of the previous frame. k-1 Calculate the prediction state for the current frame: ;

[0074] Synchronously update the state covariance matrix P k Q is the process noise covariance matrix, preset according to the system motion characteristics:

[0075] ;where F k Let x be the state transition matrix. k-1 The state vector of the previous frame, Δt is the inter-frame time interval (in seconds). .

[0076] S3.2.2, Using the Kalman filter prediction position as the initial search starting point for the optical flow method, calculate the displacement within a small window near the prediction point and execute the following: ;

[0077] I x I y For spatial gradient, I t This represents the time gradient.

[0078] S3.2.3 Template matching for refined positioning: To further improve tracking accuracy, the system uses a template matching algorithm to refine the results of the optical flow method.

[0079] 1) In the previous frame image, a fixed-size template region T is extracted with the original monitoring point as the center;

[0080] 2) In the current frame image, extract the search region I centered on the optical flow tracing position;

[0081] 3) Template matching using normalized cross-correlation coefficient (NCC):

[0082] ;in:

[0083] T(x,y): Pixel values ​​in the template region;

[0084] I(x,y): Pixel values ​​of the search region;

[0085] , The template and search area mean values; in the matching results, the position corresponding to the maximum value of NCC is the corrected monitoring point position.

[0086] If the maximum NCC value is greater than or equal to the set threshold (e.g., 0.85), the match is considered valid, and that position is adopted as the final point.

[0087] S3.2.4 When the matching confidence level is below 0.85, enable forward and backward optical flow consistency check to avoid drift:

[0088] 1) Calculate the optical flow backward using the refined point p2 of the current frame to obtain the position estimate of the previous frame. ;

[0089] 2) If the reverse deviation (typically δ=2 pixels), indicating that a tracking jump may occur;

[0090] 3) At this point, the optical flow result is not adopted, and the Kalman filter prediction value in S3.1 is used as the final point.

[0091] Kalman-optical flow joint tracking has, to some extent, solved the problem of inaccurate tracking of monitoring points caused by changes in illumination and local occlusion.

[0092] S3.3 The calculation of the distance and orientation angle changes between the reference point and the monitoring point is as follows:

[0093] During feature point tracking, the system extracts the pixel coordinates (x, y, y) of the monitoring points frame by frame from the video frame. t ,y t ) and the corresponding reference point pixel coordinates (x r0 ,y r0 It calculates the changes in distance and orientation angle between the current image monitoring point and the initial image reference point in real time.

[0094] The pixel distance change Δd is expressed as:

[0095] ;

[0096] The pixel orientation angle change Δθ is expressed as: ;

[0097] Where (x) t ,y t (x) represents the pixel coordinates of the monitoring point at time t in the current frame. r0 ,y r0 () represents the pixel coordinates of the reference point in the initial frame.

[0098] The system compares the aforementioned distance change Δd with the calibrated pixel displacement threshold T. d，pix The change in orientation angle Δθ and the calibrated pixel orientation angle threshold T θ，pix The comparison is performed; if any one of the indicators exceeds the corresponding threshold, it is considered that there is a sign of landslide or collapse at that monitoring point.

[0099] In step S3, when a monitoring object has multiple reference points, the spatial relationship between the monitoring point and its corresponding set of reference points is evaluated. The system calculates the distance change Δd and the direction angle change Δθ between the monitoring point and each reference point. The above multiple sets of distance changes and direction angle changes are fused. The fusion method includes, but is not limited to, the maximum value and the average value.

[0100] In step S3, the signs of collapse and toppling at the monitoring point include, but are not limited to, the maximum displacement, the average displacement, the maximum change in orientation angle, and the average change in orientation angle.

[0101] Specifically, step 4 includes:

[0102] S4.1 Assess the collapse and toppling situation of each monitored object: Determine the risk level based on the maximum displacement and average displacement of the monitoring points of each monitored object;

[0103] S4.2 Assess the overall landslide and collapse trend: Determine the overall landslide and collapse trend based on the weighted risk level of each monitored object.

[0104] In step S4, the risk level of a single monitored object includes, but is not limited to, the volume of landslides and collapses at the monitoring point, the maximum displacement, the average displacement, the maximum change in direction angle, and the average change in direction angle; the determination of the overall landslide and collapse risk, i.e., the alarm level, includes, but is not limited to, the weighted risk level, the average risk level, and the maximum risk level of the monitored object.

[0105] Specifically, step 5 includes:

[0106] S5.1 Abnormal Target Judgment and Alarm. Based on the displacement or change in direction angle, the system automatically compares the set multi-level alarm levels (e.g., levels 1-9, corresponding to different audible and visual signals) and triggers the audible and visual alarm devices. The alarm level is proportional to the area's risk level; higher risk levels trigger stronger and more frequent audible and visual signals to alert on-site personnel.

[0107] S5.2. Once all monitoring points have moved out of the monitoring screen, stop tracking the monitoring points and stop the alarm.

[0108] S5.3 Alarm Information Output. During the monitoring point tracking process, the coordinates of the monitoring point in each frame of the video stream are recorded and saved in the log file for retrospective analysis of landslide precursors.

[0109] This invention proposes a landslide and collapse alarm method based on monitoring video streams and distance-direction angle relationships. It continuously collects video streams of the monitored area using a fixed camera, combines these with user-selected reference points and monitoring points within the images, and constructs the distance and direction angle relationships between the reference and monitoring points for dynamic monitoring and alarm assessment. The reference points are prominent feature points in stable areas of the monitoring images (such as exposed rock masses or fixed structures), while the key monitoring points are typical feature points in critical areas of potential landslide and collapse zones, such as rock cracks and block edges. The changes in distance and direction angle between these key monitoring points and the reference points are calculated frame-by-frame.

[0110] This invention utilizes changes in the spatial structure (distance, orientation angle) between monitoring points and reference points to reflect the potential movement trend of a sliding body. Compared to traditional methods, it does not rely on physical sensors or physical contact, achieving pixel-level monitoring point tracking and real-time identification of collapse and toppling trends. It boasts advantages such as high frequency, low latency, near real-time performance, and wide coverage. Furthermore, by constructing a weighted correlation model between monitoring points and their respective monitored objects, and evaluating the overall collapse and toppling situation, it enables multi-target parallel analysis and hierarchical alarms, significantly improving the timeliness and reliability of collapse and toppling alarms.

[0111] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions and substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A landslide and collapse alarm method based on the relationship between monitoring video stream and distance-direction angle, characterized in that, A landslide and collapse alarm system is used for real-time monitoring and alarm of potential landslide and collapse areas. The landslide and collapse alarm system includes a monitoring stream acquisition device, a video stream processing module, a distance-direction angle relationship calculation module, and an alarm output module, which are used to realize a closed-loop process of monitoring video stream acquisition, processing, anomaly detection, and alarm output. The landslide and collapse alarm method includes the following steps: S1. Acquire real-time video streams for monitoring; S2. Calibrate the camera parameters and select reference points and key monitoring points for each monitored object; S3. Dynamically track and monitor points and analyze the slippage situation; S4. Analyze the slippage trend of each monitored object and the overall slippage trend based on the slippage trend of the monitoring points; S5. Determine the landslide warning level and issue an audible and visual alarm; In step S3, the monitoring point tracking algorithm used when dynamically tracking the monitoring point includes a dynamic coupling algorithm that combines Kalman filtering and optical flow: the monitoring point position of the current frame is predicted according to Kalman filtering, and the predicted position is used as the initial search point of optical flow; and the monitoring point coordinates are further corrected using template matching method. In the dynamic coupling algorithm combining Kalman filtering and optical flow, specifically: in the previous frame image, a fixed-size template region T is extracted centered on the original monitoring point; in the current frame image, a search region I is extracted centered on the optical flow tracing position; template matching is performed using the normalized cross-correlation coefficient (NCC): the position corresponding to the maximum NCC value in the matching result is the corrected monitoring point position; if the maximum NCC value is greater than or equal to a set threshold of 0.85, the matching is considered valid, and this position is adopted as the final point; when the matching confidence is lower than 0.85, a forward and backward optical flow consistency check is enabled to avoid drift: the optical flow is calculated backward using the refined point p2 of the current frame to obtain the position estimate of the previous frame. If the reverse deviation If δ=2 pixels, it indicates that a jump may occur during tracking; in this case, the optical flow result is not adopted, and the Kalman filter prediction value is used as the final position.

2. The landslide and collapse alarm method according to claim 1, characterized in that, In step S1, a monitoring stream acquisition device is deployed in the safe and stable zone opposite the potential landslide and collapse zone to acquire monitoring video streams in real time and transmit them to the video stream processing module. The device used to acquire the video streams is a monitoring camera, which includes a high-definition CCD monitoring camera and an infrared camera.

3. The landslide and collapse alarm method according to claim 1, characterized in that, In step S2, the user needs to input basic parameters of the monitoring area. Based on these basic parameters, the landslide and collapse alarm system uses the principle of similar triangles to construct a mapping relationship between the actual terrain displacement and the pixel offset in the image. This mapping relationship is used to compare the pixel-level changes of key points relative to reference points in subsequent images to determine whether there are signs of landslides or collapses at the monitoring points.

4. The landslide and collapse alarm method according to claim 1, characterized in that, In step S2, the reference point is a relatively stable and unchanging area in the image, including exposed rock, pile foundations, and fixed structures in the stable area; the monitoring point is a key feature point in the area with significant risk of collapse and toppling, including rock mass cracks, convex points of hidden danger bodies, and block edges.

5. The landslide and collapse alarm method according to claim 1, characterized in that, In step S3, the spatial relationship change of each monitoring point relative to the reference point is calculated in real time, specifically including: In the feature point tracking process, the pixel coordinates (x, y, y) of the monitoring point are extracted frame by frame from the video frame. t ,y t ) and the corresponding reference point pixel coordinates (x r0 ,y r0 ), calculates the changes in distance and orientation angle between the current image monitoring point and the initial image reference point in real time; The pixel distance change Δd is expressed as: ； The pixel orientation angle change Δθ is expressed as: ； Where (x) t ,y t (x) represents the pixel coordinates of the monitoring point at time t in the current frame. r0 ,y r0 () represents the pixel coordinates of the reference point in the initial frame; The above distance change Δd is compared with the calibrated pixel displacement threshold T. d，pix The change in orientation angle Δθ and the calibrated pixel orientation angle threshold T θ，pix The comparison is performed; if any item exceeds the corresponding threshold, it is considered that there is a sign of landslide or collapse at that monitoring point.

6. The landslide and collapse alarm method according to claim 1, characterized in that, In step S3, when there are multiple reference points for a certain monitoring object, the spatial relationship between the monitoring point of the monitoring object and its corresponding set of reference points is evaluated. The system calculates the distance change Δd and the direction angle change Δθ between the monitoring point and each reference point respectively. The above multiple sets of distance changes and direction angle changes are fused, and the fusion method includes the maximum value and the average value. In step S3, the signs of collapse and toppling at the monitoring point include the maximum displacement, average displacement, maximum change in orientation angle, and average change in orientation angle.

7. The landslide and collapse alarm method according to claim 1, characterized in that, In step S4, the factors for determining the risk level of a single monitored object include the volume of landslide and collapse at the monitoring point, the maximum displacement, the average displacement, the maximum change in direction angle, and the average change in direction angle; the factors for determining the overall landslide and collapse risk, i.e., the alarm level, include the weighted risk level, the average risk level, and the maximum risk level of the monitored object.

Images