Falling stone detection and early warning method and system fusing image recognition and behavior judgment

Abstract

The invention relates to the technical field of video image intelligent analysis, in particular to a rockfall detection and early warning method and system based on videos. According to the method, a suspected rockfall target is recognized and judged by collecting a road or side slope monitoring video and combining target detection and motion trail analysis; and after the rockfall is judged, generating early warning information containing the position, time and danger level, storing the early warning information in a database, and triggering an early warning prompt of a user side at the same time. The system is composed of a video acquisition module, a detection and judgment module, an early warning generation module, a log storage module and a front-end interaction module, and all the modules achieve cooperation through data interfaces. Compared with the prior art, the method can still keep high detection accuracy and real-time performance in a complex environment, has the advantages of high automation degree and high expandability, can be widely applied to scenes such as highways, railways and mining areas, and remarkably improves the traffic and engineering safety level.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and early warning technology, and in particular to a method and system for detecting and warning of falling rocks that integrates image recognition and behavior determination. This method and system are applicable to scenarios such as highways, railways, and mining slopes, and can also be extended to various risk environments such as dams, tourist trails, and geological disaster monitoring points. Background Technology

[0002] Current rockfall monitoring mainly relies on manual inspections or methods based on single-image detection. Manual inspections are inefficient, have significant time lags, and are difficult to detect potential rockfall risks in a timely manner. Furthermore, manual inspections pose significant safety hazards in severe weather or dangerous terrain.

[0003] Existing video-based rockfall detection methods mostly perform static analysis on single-frame images, lacking the tracking and judgment of the cross-frame motion trend of suspected targets. This leads to false alarms and missed alarms under complex backgrounds, lighting changes, or partial occlusion conditions, resulting in insufficient detection stability.

[0004] Furthermore, existing systems typically lack dynamic early warning mechanisms based on risk levels, failing to create a closed loop between detection results and response measures. On one hand, detection results lack multi-dimensional parameters such as speed, acceleration, and directional consistency, making it difficult to comprehensively reflect the target's risk level. On the other hand, early warning linkage functions are limited, failing to achieve real-time control with traffic facilities or operational systems, resulting in untimely and ineffective early warning responses. Summary of the Invention

[0005] The purpose of this invention is to address the problems of low efficiency of manual inspection, insufficient accuracy of detection based on single-frame images, and lack of risk classification and early warning mechanisms in the existing technology, and to provide a rockfall detection and early warning method and system that integrates image recognition and behavior judgment, so as to achieve high-precision detection, dynamic behavior judgment and multi-level risk early warning of rockfall events.

[0006] The technical problem this invention aims to solve is that existing rockfall monitoring methods have the following shortcomings:

[0007] First, the detection accuracy is insufficient, especially under complex backgrounds, lighting changes, and occlusion conditions, which can easily lead to false alarms and missed alarms.

[0008] Secondly, there is a lack of analysis of the movement trends and behavior of suspected falling rock targets, and static identification can only be performed based on a single frame image, resulting in an incomplete risk assessment.

[0009] Third, the early warning mechanism is simplistic and lacks risk level classification based on multi-dimensional characteristics, failing to achieve a closed loop throughout the entire process from detection and judgment to coordinated response.

[0010] To address the aforementioned problems, this invention proposes a method for detecting and warning of falling rocks. The method includes: acquiring a monitoring video stream via a camera and performing image preprocessing; detecting suspected falling rock targets in the preprocessed images and generating candidate targets; performing cross-frame tracking and motion trend analysis on the candidate targets; classifying the candidate targets into risk levels based on parameters such as velocity, acceleration, and directional consistency; and generating corresponding warning information according to the risk level and outputting it to a terminal or linkage response device.

[0011] This invention also proposes a rockfall detection and early warning system, which includes a video acquisition module, a video preprocessing module, a detection core module, a model and parameter management module, a hazard level assessment module, an early warning generation module, and a user interaction module. All modules are connected via a data bus and uniformly scheduled by a control unit, achieving full-process processing and early warning linkage of the monitoring video.

[0012] The video acquisition module is used to acquire monitoring video streams of roads or slopes; the video preprocessing module is used to perform operations such as jitter reduction, illumination compensation, and region cropping on video frames; the detection core module identifies suspected rockfall targets based on deep learning detection networks or inter-frame difference methods; the model and parameter management module is used to store detection models and threshold parameters and supports dynamic updates; the hazard level assessment module classifies target risks based on characteristics such as speed, acceleration, and directional consistency; the early warning generation module generates corresponding early warning information; and the user interaction module transmits early warning results to terminals or triggers external linkage devices.

[0013] Through the coordinated operation of the above modules, this invention can complete the entire process from video acquisition, image processing, target detection to risk assessment and early warning output, ensuring the accuracy of detection results and the timeliness of early warning response.

[0014] In a preferred embodiment, the method and system of the present invention can accurately determine potential rockfall behavior by cross-frame tracking and motion trend analysis of suspected rockfall targets, combined with multi-dimensional feature parameters such as speed, acceleration, directional consistency and target size coverage; and by constructing a risk scoring function and a dual threshold determination mechanism, rockfall risk is divided into three levels: low risk, medium risk and high risk, thereby achieving graded early warning.

[0015] Furthermore, the system of this invention can generate corresponding early warning information based on the risk level. Low-risk events are only recorded; medium-risk events are pushed to terminal devices through the user interaction module; high-risk events trigger audible and visual alarms or traffic control linkage measures, realizing real-time alarm and response linkage.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) By combining video acquisition, image preprocessing and deep learning detection model, the accuracy and robustness of rockfall detection are significantly improved;

[0018] (2) By introducing cross-frame tracking and motion trend analysis, the downward trend of falling rocks can be effectively determined, reducing misjudgments caused by static detection;

[0019] (3) Through risk level classification and dual threshold judgment mechanism, the risk of rockfall is classified and early warning is realized, which ensures the accuracy and operability of the early warning information;

[0020] (4) Through the linkage between the early warning generation and user interaction modules, multi-level response strategies are supported, forming a complete closed loop from simple recording to linkage control;

[0021] (5) The system is modular and scalable, and can adapt to various application scenarios such as roads, railways, mining areas, dams and tourist attractions, and has a wide range of practical value.

[0022] In summary, this invention has achieved significant improvements in detection accuracy, behavior determination capability, risk classification and early warning mechanism, and application scope. It can effectively overcome the shortcomings of existing technologies and provide reliable protection for road traffic safety, railway transportation safety, and mining operation safety. Attached Figure Description

[0023] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly described below. Obviously, the following drawings are merely some schematic diagrams of the present invention, used to illustrate the basic concept of the present invention. Those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the method flow described in Embodiment 1 of the present invention, used to illustrate the basic steps of the rockfall detection and early warning method, but is not limited to this embodiment.

[0025] Figure 2 This is a schematic diagram of the system structure described in Embodiment 2 of the present invention, used to illustrate the relationship between the modules, but is not limited to this embodiment.

[0026] Figure 3 This is a schematic diagram of the three-layer interaction architecture described in Embodiment 2 of the present invention, used to illustrate the data interaction between the user side, the system core, and the management side, but is not limited to this embodiment.

[0027] Figure 4 This is a schematic diagram of the lifecycle state machine according to Embodiment 3 of the present invention, used to illustrate the state transition logic of the candidate target, but is not limited to this embodiment.

[0028] Figure 5 This is a schematic diagram of the hazard level determination logic described in Embodiment 4 of the present invention, used to illustrate the risk assessment process after multi-dimensional feature fusion, but is not limited to this embodiment.

[0029] Figure 6 This is a schematic diagram of the application scenario described in Embodiment 5 of the present invention, used to illustrate the deployment method of the system in environments such as roads, railways and mining areas, but is not limited to this embodiment.

[0030] Explanation of reference numerals in the attached figures:

[0031] 1 — Slope, 2 — Road or railway crossing, 3 — Camera location, 4 — Monitoring area, 5 — Rockfall and slide trajectory, 6 — Detection and early warning system, 21 — Video acquisition module, 22 — Video preprocessing module, 23 — Core detection module, 24 — Model and parameter management module, 25 — Hazard level assessment module, 26 — Early warning generation module, 27 — User interaction module Detailed Implementation

[0032] The specific embodiments of the present invention will now be described with reference to the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Equivalent modifications made by those skilled in the art without departing from the spirit and substance of the present invention should be considered within the scope of protection of the present invention.

[0033] like Figure 1 As shown, one embodiment of this application provides a rockfall detection and early warning method, which includes steps S100 to S600.

[0034] Step S100: The camera captures the monitoring video stream and decodes it to obtain a frame-by-frame image sequence.

[0035] Step S200: Perform jitter removal, illumination compensation, and region cropping on the frame-by-frame images to obtain a preprocessed image for detection.

[0036] The purpose of step S300 is to effectively identify potential rockfall targets in the preprocessed image and remove background interference, thereby providing reliable input for subsequent motion analysis.

[0037] Step S300: Detect suspected rockfall targets in the preprocessed image and output candidate targets and their confidence scores. For example... Figure 1 As shown, this step corresponds to the candidate target detection module. This step includes:

[0038] Step S301: Receive the preprocessed image, define the detection area based on scene calibration, and reduce interference from irrelevant areas;

[0039] Step S302: Generate candidate targets using an object detection network or inter-frame difference method, and output bounding boxes and confidence scores;

[0040] Step S303: Perform non-maximum suppression on overlapping candidates and merge redundant targets;

[0041] Step S304: Filter candidates based on confidence threshold and size range, and eliminate noisy targets;

[0042] Step S305 generates structured records (including timestamps, bounding box positions, target confidence, etc.) for the filtered candidate targets and passes them to the behavior determination step for cross-frame tracking and trend analysis.

[0043] Step S400 involves cross-frame tracking and motion trend analysis of the candidate target to determine if a downward trend exists. This step includes:

[0044] Step S401: The trajectory is maintained by inter-frame IoU matching and Kalman filtering to handle target survival / death and occlusion.

[0045] Step S402: Calculate the instantaneous velocity and acceleration of the candidate target based on its centroid position in adjacent frames.

[0046]

[0047]

[0048] in Indicates the inter-frame time interval. Indicates speed, This represents acceleration. The calculation result is used for subsequent motion trend determination.

[0049] The velocity and acceleration calculations are not only used for single-frame analysis, but also serve as important input parameters for subsequent trend determination and risk scoring.

[0050] Step S403: Based on the slope calibration, statistically analyze the consistency of the target direction and the satisfaction of the threshold.

[0051] Step S404: When a candidate target satisfies the following combination of conditions, it is determined that it has a downward trend:

[0052] (1) In the most recent consecutive frames, at least some frames have velocity and acceleration that simultaneously satisfy the threshold condition;

[0053] (2) The percentage of frames whose motion direction is consistent with the slope calibration exceeds the preset percentage;

[0054] (3) The continuity of the target trajectory meets the minimum number of continuous frames requirement.

[0055] If all the above conditions are met, the target enters the hazard level assessment step; if not all conditions are met, the target record is terminated, and its status is determined to be 'lost' or 'fallen'.

[0056] Step S500 involves classifying the risk level of the target based on characteristics such as velocity, acceleration, directional consistency, scale coverage, and confidence level. This step includes:

[0057] Step S501, construct feature vectors Normalize to [0,1] by interval mapping.

[0058] Step S502: Construct a feature vector containing parameters such as velocity, acceleration, directional consistency, target size coverage, and detection confidence, and obtain a comprehensive risk score using a weighted summation method.

[0059]

[0060] The weights of each parameter can be obtained through scenario calibration or historical data statistics to ensure the rationality and adaptability of the risk score. The weights of each parameter can be dynamically adjusted based on the application scenario through historical statistics or machine learning training.

[0061] Step S503: Output the level according to the double threshold rule:

[0062]

[0063] in The preset threshold can be set according to the scenario or historical statistics.

[0064] Step S504: Output the judgment result to the early warning module. The output content shall include at least the target number, location, timestamp, risk level and related parameters, so that the early warning generation module can further process it.

[0065] Step S600: Generate early warning information based on risk level: low risk is only recorded, medium risk is pushed to the terminal, and high risk triggers sound and light or linkage measures and saves complete records.

[0066] In another embodiment, such as Figure 2 As shown, the rockfall detection and early warning system includes a video acquisition module 21, a video preprocessing module 22, a detection core module 23, a model and parameter management module 24, a hazard level assessment module 25, an early warning generation module 26, and a user interaction module 27. These modules are connected via a data bus and coordinated by a control unit to form a complete detection and early warning functional chain.

[0067] like Figure 2As shown, this embodiment embodies a modular implementation of the method steps in its system architecture. Specifically, the video acquisition module 21 is used to acquire monitoring video streams of roads or slopes and transmit the data to the video preprocessing module 22 in real time; the latter performs jitter removal, illumination compensation, region cropping, and noise filtering on the input video frames to obtain preprocessed images suitable for detection. The detection core module 23 receives the preprocessing results and outputs the bounding boxes, categories, and confidence information of suspected rockfall targets based on deep learning detection networks or background subtraction methods.

[0068] The detection results are transmitted to the hazard level assessment module 25. The hazard level assessment module 25 comprehensively scores the target's characteristics, including speed, acceleration, directional consistency, scale coverage, and confidence level, and classifies the risk level according to preset thresholds. The risk level results are then transmitted to the early warning generation module 26.

[0069] The warning generation module 26 generates corresponding warning information based on the hazard level and outputs it externally through the user interaction module 27. The user interaction module 27 may include a display terminal, mobile device application, or traffic management interface for receiving and presenting the warning information. When a high-risk situation is determined, the user interaction module 27 can also link with external facilities, such as triggering audible and visual alarm devices or controlling traffic signal equipment, thereby achieving real-time warning and coordinated response.

[0070] In this embodiment, steps S100 to S600 in Method Embodiment 1 are all implemented by the aforementioned modules: steps S100 and S200 are completed by the video acquisition module 21 and the video preprocessing module 22; steps S300 and S400 are completed by the detection core module 23 with the support of the model and parameter management module 24; step S500 is executed by the hazard level assessment module 25; and step S600 is executed by the early warning generation module 26 and the user interaction module 27. This establishes the correspondence between the method steps and the system modules.

[0071] In another embodiment, such as Figure 3 As shown, the system can be divided into three parts: the user side, the system core, and the management side. The user side includes real-time monitoring, video uploading, and early warning reception functions; the system core includes video access, detection and judgment, and early warning services; and the management side includes device management, parameter configuration, and interface integration. These three components interact via network connections and data interfaces, forming an end-to-end data and control flow. Monitoring videos uploaded by the user side are detected and risk-judged by the system core, and early warning results are displayed on the user side, triggering responses. The management side can remotely manage and adjust model parameters, threshold configurations, and device operating status.

[0072] In summary, this embodiment achieves this through... Figure 2 The modular structure shown is Figure 3The three-layer interactive architecture shown achieves a unified approach to methodological processes, system structure, and functional interaction. This system not only completes each step of the rockfall detection and early warning method but also establishes stable data and control flows between the user side, the system core, and the management side, thereby ensuring the integrity and scalability of the detection and early warning process.

[0073] like Figure 4 As shown, this embodiment provides a candidate target lifecycle management mechanism. Unlike the linear step-based description in Embodiment 1, this embodiment uses a state machine approach to model candidate targets, more intuitively revealing the dynamic evolution of targets during detection, judgment, and early warning processes.

[0074] like Figure 4 As shown, the lifecycle state machine characterizes the entire process of a candidate target from detection to warning through state transitions, making the logic more rigorous. In another embodiment, this invention proposes a lifecycle management mechanism for candidate targets. This mechanism models each stage of the target monitoring process using a state machine, giving detection, judgment, and warning clear logical relationships and traceability. Compared with the linear step description in Embodiment 1, this embodiment uses a state transition method, which more intuitively reflects the evolution of candidate targets under different conditions.

[0075] When no suspected rockfall target is detected within the monitored area, the system is in an "undetected" state.

[0076] Once the confidence level output by the core detection module is greater than the initial threshold... Once the bounding box is defined, the target enters the "candidate" state.

[0077] In the candidate state, if the target is in continuous Detected in the frame, with average confidence. satisfy ,

[0078] The target then enters the "tracked" state; if the condition is not met, it returns to the "undetected" state.

[0079] In the “tracked” state, the system maintains the target trajectory and calculates the velocity and acceleration, and the calculation method is shown in formula (1) and (2).

[0080] When in recent Within the frame, the target simultaneously satisfies both velocity and acceleration threshold conditions: ,

[0081] When the direction consistency satisfies the constraint shown in formula (3), the target enters the "slide confirmation" state; if the trajectory is interrupted or the condition is not met, it enters the "fallback" or "loss" state.

[0082] In the "slide confirmation" state, the system fuses the multi-factor features of candidate targets and calculates the event comprehensive factor. :

[0083]

[0084] in, and These are velocity and acceleration (see formulas (1) and (2)). To achieve the target scale coverage, To test the confidence level, ~ These are weighting coefficients, and they satisfy... .

[0085] then The target will enter the "risk assessment" state; otherwise, it will remain in the "tracked" state, awaiting further observation.

[0086] In the "risk assessment" state, the hazard level assessment module constructs a feature vector and calculates the comprehensive score according to formula (4). The target status is then classified according to formula (5). If the result is medium risk, the target status is changed to "warning"; if it is high risk, the target status is changed to "alarmed".

[0087] In "Warning" mode, the system generates a warning message and pushes it to the terminal via the user interaction module to alert operators or drivers to the risks. In "Alarmed" mode, the system marks the target as a high-risk event and triggers audible and visual alarms or traffic-related measures, such as closing lanes or controlling traffic signals. All such events are recorded and archived in the database.

[0088] In another embodiment, some states in the target lifecycle state machine can be merged or pruned according to actual needs. For example, the "candidate" and "tracked" states can be merged into a single "tracked" state to reduce the number of states; alternatively, the "warning" and "alarmed" states can be merged into an "alarm" state, where different levels can be distinguished by a risk level field. For scenarios with limited computing resources, the "swipe down to confirm" state can be omitted, and the process can directly proceed from "tracked" to "risk assessment." All of the above adjustments fall within the scope of protection of this invention. Figure 4 As shown, the state machine realizes the evolution of the candidate objective at different stages.

[0089] In summary, such as Figure 4The lifecycle state machine shown defines the complete transition logic from detection to alarm for a candidate target, and introduces decision conditions and comprehensive factor calculations at key nodes. Compared with the linear process in Embodiment 1, this embodiment more intuitively reveals the evolution process of the target at different stages through state transition logic, making the detection and early warning logic of this invention more rigorous and systematic.

[0090] In another embodiment, such as Figure 6 As shown, the rockfall detection and early warning system of this invention can be applied to typical scenarios such as roads, railways, and mining areas. The system forms a complete closed loop from on-site monitoring to information output through camera deployment, video acquisition and preprocessing, target detection, risk assessment, and early warning linkage.

[0091] In road scenarios, cameras are deployed on the slopes or shoulders of mountain roads to capture real-time video streams covering potential rockfall areas on both sides of the road. When the detection result is assessed as low to medium risk, the system pushes a prompt to the variable message sign (VMS) via the user interaction module to guide vehicles to slow down. When the detection result is high risk, in addition to triggering an audible and visual alarm, the system can also directly link with traffic signal control equipment to switch the traffic lights to red, preventing vehicles from entering dangerous sections of the road, thereby avoiding traffic accidents and ensuring safe passage on mountain roads.

[0092] In railway scenarios, cameras are installed near railway slopes, tunnel entrances, or bridges to monitor potential rockfall areas around the tracks. When the detection and risk assessment module determines a target to be high-risk, the user interaction module immediately sends an alarm message to the railway dispatch center via the communication interface and triggers the trackside signal to display a red light, forcing the train to slow down or make an emergency stop. For medium-risk events, the system automatically pushes the detection results to the railway inspection personnel's terminal for manual verification and dispatching decisions, ensuring the continuity of railway transportation safety and the safety of train operation.

[0093] In the mining area, cameras are deployed on the slopes and at the entrances of transport channels to monitor in real time for potential rockfall risks that could affect mine truck operations. When the detection result is medium risk, the system pushes a risk alert to the operator's terminal through the user interaction module and highlights the risk area on the monitoring interface. When the detection result is high risk, the system can automatically close the transport channel gates, suspend mine truck operations, and trigger audible and visual alarms and personnel evacuation procedures, thereby reducing the harm of sudden accidents in the work area. All data from the entire process is fully recorded and archived for subsequent traceability and safety assessment, improving the safety protection capabilities of personnel and equipment in the work area.

[0094] exist Figure 6In the application diagram, 1 represents a slope, 2 represents a road or railway crossing, 3 represents the camera location, 4 represents the monitoring area, 5 represents falling rocks and their trajectory, and 6 represents the detection and early warning system. This diagram is used to illustrate the deployment method and area of ​​operation of the system in typical scenarios.

[0095] In another embodiment, the system of the present invention can also be applied to environments such as hydropower station dams, tourist attraction trails, and geological disaster monitoring points. In the above scenarios, the system achieves real-time monitoring and alarm for rockfalls, landslides, or other geological risks through the same data collection, detection, and early warning process, thus possessing wider applicability.

[0096] In summary, this embodiment, combining typical application scenarios of roads, railways, and mining areas, illustrates the deployment methods and risk linkage mechanisms of the system of the present invention in different environments. Specifically, the road scenario focuses on vehicle driving safety, emphasizing timely intervention through traffic lights and information screens to "improve driving safety"; the railway scenario prioritizes train operation safety, highlighting the linkage between dispatch center alarms and trackside signals to "ensure train operation safety"; and the mining area scenario aims at the safety of personnel and equipment in the work area, emphasizing the linkage between transportation channel gate control and personnel evacuation to "ensure the safety of personnel and equipment in the work area." These differentiated designs demonstrate that the present invention can not only adapt to multiple typical scenarios but also possesses good applicability and scalability, thereby ensuring the effectiveness and integrity of the system in practical applications.

[0097] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. It should be understood that the embodiments described are only for illustrating the principles of the present invention and are not intended to limit the scope of protection. Various equivalent substitutions and modifications made by those skilled in the art without departing from the spirit and essence of the present invention should be considered as included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

[0098] In summary, this invention achieves real-time detection and tiered early warning of falling rocks by integrating image recognition and behavior determination methods, combined with a modular system architecture and lifecycle management mechanism. This invention not only boasts advantages in detection accuracy and early warning reliability, but also demonstrates good applicability and promotional value in road, railway, mining area, and other geological disaster scenarios.

Claims

1. A method for detecting and warning of falling rocks that integrates image recognition and behavior determination, characterized in that, Includes the following steps: The surveillance video stream is acquired and decoded to obtain a frame-by-frame image sequence; Preprocessing is performed on the frame-by-frame images, including jitter reduction, illumination compensation, and region cropping; Suspected rockfall targets are detected in the preprocessed image, candidate targets and their confidence scores are generated, and the candidate targets are screened and recorded. The candidate target is tracked across frames, and its velocity, acceleration, and orientation consistency are calculated to determine whether there is a downward trend. A comprehensive score is calculated based on multi-dimensional features such as speed, acceleration, directional consistency, scale coverage, and confidence level, and risk levels are classified accordingly. Corresponding early warning information is generated based on the risk level, and the early warning information is output to the user terminal according to the risk level. The output includes pushing to the terminal. The linkage response includes, but is not limited to, triggering audible and visual alarms and linkage response devices in the event of a high-risk event.

2. The method according to claim 1, characterized in that, The candidate target detection module employs a convolutional neural network detection model and can combine inter-frame differencing or background modeling methods as redundant detection channels to improve the robustness and reliability of the detection results. The detection model supports online updates.

3. The method according to claim 1, characterized in that, The risk level classification includes: normalizing the multidimensional feature vector; performing weighted summation according to preset weights to obtain a risk score; and classifying it into three levels—low, medium, and high—based on a dual threshold rule to reduce false alarms and improve the accuracy of early warnings, thereby enhancing the overall reliability of the system.

4. A rockfall detection and early warning system integrating image recognition and behavior determination, characterized in that, include: The video acquisition module is configured to acquire surveillance video streams. The video preprocessing module is configured to perform preprocessing operations on video frames; The detection core module is configured to detect suspected rockfall targets and generate candidate targets; the model and parameter management module is configured to store and update the detection model and threshold parameters; the hazard level assessment module is configured to calculate a comprehensive score based on multi-dimensional features and classify risk levels, and output the risk level results for use by the early warning generation module. The output includes risk level data and related parameters. The early warning generation module is configured to generate early warning information based on the risk level. The user interaction module is configured to push early warning information to the terminal and link with external response devices in the event of a high-risk incident.

5. The system according to claim 4, characterized in that, The user interaction module includes a display terminal, a mobile terminal application, or a traffic management interface, which is configured to control traffic lights, audible and visual alarm devices, or external traffic control facilities in the event of a high-risk incident.

6. The system according to claim 4, characterized in that, The hazard level assessment module normalizes the input features and obtains a risk score based on weighted summation. The risk score is further mapped to a risk level through a dual threshold rule.

7. The system according to claim 4, characterized in that, The detection core module and the hazard level assessment module together realize the candidate target lifecycle state machine. The state machine includes states such as undetected, candidate, tracked, sliding confirmation, risk assessment, warning, and alarmed. It performs state transitions when preset conditions are met to ensure the logical consistency and traceability of the detection, judgment, and warning processes, reduce misjudgments caused by target loss, improve the traceability and stability of the warning logic, and is suitable for various monitoring environments.

8. The system according to claim 4, characterized in that, The system further includes: a traffic signal linkage module configured to interact with road traffic lights and information screens; a railway linkage module configured to interact with the railway dispatch center and trackside signals; and a mining area linkage module configured to interact with transportation channel gates and personnel evacuation devices. All modules are connected to the user interaction module via data interfaces. Each module can be enabled or disabled as needed according to different application scenarios, and the activation and deactivation of the modules are managed by the control unit.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method according to any one of claims 1 to 3.

10. The storage medium according to claim 9, characterized in that, When the detection result indicates a high risk, the program automatically triggers external collaborative response measures and saves complete event data to support post-event traceability and risk assessment. This complete event data includes detection parameters, risk level, and response records for post-event traceability and safety assessment.

Images