False alarm suppression and repeated alarm elimination method and system for monitoring power transmission line engineering vehicle

Abstract

The present application belongs to the technical field of power transmission line operation and maintenance intelligent monitoring, and particularly relates to a false alarm suppression and repeated alarm elimination method and system for monitoring of power transmission line engineering vehicles. The method and system solve the problems of high false alarm rate, frequent repeated alarms, high risk of missed alarms and poor real-time performance of the system caused by direct alarm triggering by single frame detection, insufficient adaptability of traditional matching algorithms to changes in the shape of engineering vehicles, lack of category constraints and designated type control in the existing monitoring system. The technical solution includes: collecting real-time images with point position numbers and time stamps; identifying engineering vehicles using a lightweight target detection model, and filtering and selecting effective target frames through confidence filtering; performing cross-frame tracking and unique ID assignment on consecutive frame targets based on the intersection area ratio algorithm; executing alarm rules, reporting the first frame of the specified vehicle type target of the new tracking ID, triggering re-reporting and clearing historical records when the reported ID meets the change conditions; and updating the point position number of the effective target information to the historical scene database.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for power transmission line operation and maintenance, and provides a method and system for suppressing false alarms and eliminating repeated alarms in the monitoring of power transmission line engineering vehicles. Background Technology

[0002] Publicly known technologies:

[0003] The illegal operation of construction vehicles (excavators, cranes, pile drivers, etc.) within the protection zone of power transmission lines is a core cause of power line tripping, equipment damage, and power accidents. The industry generally adopts a technical solution that combines camera visual monitoring with target detection algorithms to monitor the intrusion of construction vehicles. This involves collecting real-time images through monitoring cameras deployed along the power transmission line, using a deep learning target detection model to identify construction vehicles in the images, and directly triggering an alarm when a target is detected, pushing the alarm information to the terminal of the operation and maintenance personnel to achieve timely early warning of construction vehicle intrusion.

[0004] Existing technical solutions:

[0005] The core execution process of the existing engineering vehicle monitoring system is as follows: the monitoring camera continuously collects image data within the power transmission line protection zone → the image data is transmitted to the edge computing device or cloud server → a general target detection model (such as YOLO series, Faster R-CNN, etc.) is used to identify engineering vehicles in a single frame image → if an engineering vehicle target is detected, an alarm message is immediately generated and pushed → maintenance personnel conduct on-site verification and handling based on the alarm message.

[0006] Some optimization schemes only improve model accuracy in the target detection stage, without designing an alarm judgment stage after detection, and still directly trigger alarms based on single-frame detection results; moreover, existing tracking schemes generally use IOU (Intersection over Union) as the target matching basis, which has low sensitivity to morphological changes of engineering vehicles during operation, and is prone to tracking errors or missed alarm judgments.

[0007] Disadvantages of existing technical solutions

[0008] 1. High false alarm rate: Affected by natural environment (sudden changes in lighting, rain, snow, fog), scene interference (tree obstruction, similar objects such as farm vehicles / trucks), and equipment errors (blurred camera, image shaking), target detection of single frame images is prone to misidentifying non-engineering vehicle targets as intrusion targets, or identifying environmental debris as engineering vehicles, generating a large number of invalid false alarms and consuming the verification resources of operation and maintenance personnel.

[0009] 2. Frequent repeated alarms: The same engineering vehicle will be repeatedly identified by the detection model in consecutive frame images. The system will generate dozens or even hundreds of alarm messages for the same target in a short period of time, resulting in repeated alarms. This will cause information congestion on the operation and maintenance terminal, and the operation and maintenance personnel will not be able to quickly determine the actual intrusion situation.

[0010] 3. High risk of false negatives: Some solutions use stringent rules such as multi-frame continuous verification and historical scene verification to filter targets, which may filter out "temporarily intruding but real construction vehicles" (such as cranes passing through the protected area at high speed), resulting in false negatives; at the same time, newly added intrusion targets around the construction site may be misjudged as fixed targets due to historical scene verification rules, further increasing the false negative rate.

[0011] 4. Poor system adaptability: Some solutions use complex tracking and verification algorithms, which involve a large amount of computation and cannot be adapted to the low computing power of edge computing equipment along the transmission line, resulting in insufficient real-time performance of the monitoring system and missing the best time for handling.

[0012] 5. Dual defects in category matching and matching algorithm: The existing tracking scheme does not consider the differences in target categories, which can easily lead to mismatching engineering vehicles of different categories as the same target; at the same time, the IOU (Intersection over Union) matching algorithm used has poor adaptability to target box scaling and partial occlusion, which can easily lead to cross-category tracking errors and ID confusion due to matching algorithm defects, which can further trigger alarm errors; and there is no targeted alarm control for "specified risk vehicle types", which can easily trigger invalid alarms for non-risk vehicles. Summary of the Invention

[0013] The purpose of this invention is to solve the problems of high false alarm rate, frequent repeated alarms, high risk of missed alarms, and poor real-time performance in existing power transmission line engineering vehicle monitoring systems, caused by factors such as single-frame detection directly triggering alarms, low sensitivity of traditional IOU matching algorithms to changes in the shape of engineering vehicles, lack of category constraints and specified type control.

[0014] To achieve the above objectives, the present invention employs the following technical means:

[0015] A method for suppressing false alarms and eliminating recurring alarms in the monitoring of power transmission line engineering vehicles includes the following steps:

[0016] Step S1: Collect real-time images by using cameras deployed at various monitoring points in the power transmission line protection zone, and add a unique location number and timestamp to each frame of the image;

[0017] Step S2: Use a lightweight target detection model to identify engineering vehicles in the real-time image, output the bounding box coordinates, target category and confidence score of the target, and filter the valid target boxes through the confidence score filtering formula to obtain a list of valid detected targets;

[0018] Step S3: Based on the IOA algorithm, perform cross-frame tracking and ID assignment on valid detected target boxes in consecutive frames, where IOA is the ratio of the intersection area to the current target box area, and output a list of targets with unique tracking IDs;

[0019] Step S4: Execute alarm generation and duplicate alarm elimination rules, including: when a new tracking ID appears for the first time, if the target type belongs to the specified vehicle type, it should be reported immediately in the first frame; for a track ID that has already been reported, when the change judgment condition is met, a re-report alarm is triggered and the historical target box record of the tracking ID is deleted.

[0020] Step S5: Associate the detected valid target information with the location number and update it to the historical scene database.

[0021] In the above scheme, in step S2, a list of specified vehicle types is preset, and valid target boxes are filtered using a confidence filtering formula:

[0022]

[0023] in, The confidence score for the engineering vehicle category output by the target detection model, with a value ranging from 0 to 1; A preset confidence threshold of 0.7 is set; only target boxes with a confidence level greater than or equal to the preset confidence threshold are retained, and a list of valid detected targets containing target category information and belonging to the specified vehicle type is output.

[0024] In the above scheme, step S3, which involves cross-frame tracking and ID allocation based on the IOA algorithm, includes:

[0025] Constraints: Target boxes of different categories belong to different targets; IOA matching is only performed on target boxes of the same category and belonging to the specified vehicle type.

[0026] The formula for calculating IOA is:

[0027]

[0028] in, The target bounding box for the current frame. This is the historical target bounding box from the previous frame. For the target box With target box The area of ​​intersection For the target box The area;

[0029] IOA matching rules:

[0030] If the IOA value of the current target box and a target box of the same category in the previous frame is greater than a preset threshold, they are determined to be the same target. The tracking ID and the latest bounding box coordinates of the target are updated, and the number of undetected frames of the tracking ID is reset to zero.

[0031] If the IOA of the current target box and multiple target boxes of the same category in the previous frame are all greater than the preset threshold, then the target box with the largest IOA value is selected for matching.

[0032] If the IOA value of the current target box and all target boxes of the same category in the previous frame are less than or equal to the preset threshold, it is determined to be a new target, a unique new tracking ID is assigned to it and it is added to the tracking list;

[0033] ID cleanup rules: For tracking IDs that are not matched in the tracking list, increment the number of frames not detected by one; if the target corresponding to a tracking ID is not detected for three consecutive frames, remove it from the tracking list.

[0034] In the above scheme, the preset threshold is configured according to the scene type of the monitoring point: 0.4 for points around the construction site, 0.6 for points in the urban area, and 0.5 for points in the suburbs; and the target box is smoothed before IOA calculation, and the average coordinate of the target box in the current frame and the previous frame is taken.

[0035] In the above scheme, the change determination condition in step S4 is that the following two sub-conditions are satisfied simultaneously:

[0036]

[0037] in, k∈{1,2,3} represents the bounding box of the current frame and the bounding box of the same category in the three historical frames. The IOA value of the frame target bounding box. IOA thresholds configured for different scenarios; The Pearson correlation coefficient between the target regions of the current frame and the previous frame. This is the threshold for the Pearson coefficient.

[0038] In the above scheme, the formula for calculating the Pearson correlation coefficient is:

[0039]

[0040] in, A flattened array of grayscale values ​​of the target region in the current frame after downsampling. This is a flattened array of grayscale values ​​from the same tracking ID target region in the previous frame after downsampling. , They are respectively , The average value; the Pearson coefficient threshold is configured according to the scenario: 0.6 for urban points, 0.8 for suburban points, and 0.7 for points around construction sites.

[0041] In the above scheme, before the target change determination in step S4, the regional image is downsampled to 64×64 pixels, and only the most recent three frames are retained in the historical target box; the specified vehicle type list by default includes cranes, pump trucks, excavators, pile drivers, and bulldozers.

[0042] This invention also provides a false alarm suppression and repetitive alarm elimination system for monitoring power transmission line engineering vehicles, comprising:

[0043] The image acquisition module is used to acquire real-time images through high-definition cameras at various monitoring points in the power transmission line protection zone, and add point numbers and timestamps to the images;

[0044] The lightweight engineering vehicle detection module has a built-in pre-trained lightweight engineering vehicle target detection model and a list of specified vehicle types. It is used to identify and classify engineering vehicles in real-time images, and outputs a list of valid detection targets that contain category information and belong to the specified vehicle type through a confidence filtering formula.

[0045] The IOA tracking module has built-in IOA calculation algorithm, category matching logic, scene-specific threshold configuration module and tracking ID management logic. It only performs IOA matching on target boxes of the same category and belonging to the specified vehicle type, handles the selection logic of multiple IOA matching, and executes ID cleanup rules for three consecutive frames that have not been detected.

[0046] The alarm management module has a built-in ID alarm record dictionary, specified vehicle type verification logic, target box smoothing processing unit and change judgment unit. It only executes the alarm rule of first frame reporting plus IOA triggering change re-reporting for specified vehicle types.

[0047] The historical scene database module establishes a lightweight historical scene database for each monitoring point, storing the point number, tracking ID, target category, latest target bounding box information, and scene IOA threshold.

[0048] The main control module connects all the above modules and controls the execution order, data transmission, and interaction of each module.

[0049] In the above scheme, the number of historical target frames retained in the historical scene database module is three frames; the list of specified vehicle types can be flexibly configured through the main control module.

[0050] The present invention also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described thereon.

[0051] Compared with existing technologies whose core solutions suffer from "single-frame detection directly triggering alarms," ​​"multi-dimensional verification easily missing alarms," ​​and "poor IOU matching adaptability," this application achieves the technical effects of "reducing missed alarms, suppressing false alarms, eliminating duplicate alarms, accurate tracking, specified type control, and accurate identification of morphological changes" through architectural innovation and algorithm optimization. It also balances system real-time performance and scenario adaptability. Specific technical advantages are as follows:

[0052] 1. Effectively reduce false negative rate: Remove the stringent multi-dimensional verification module and retain only the basic confidence level filtering. When a new ID appears for the first time and belongs to a specified vehicle type, the first frame report is immediately triggered. At the same time, IOU is replaced with IOA, which increases the sensitivity to changes in the shape of engineering vehicles (crane boom, excavator boom) by more than 30%. It can effectively identify "high-risk engineering vehicles that briefly intrude but are actually high-risk" and key operation behaviors, significantly reducing false negatives.

[0053] 2. Precise control of repeated alarms: By combining the rules of "first frame reporting of specified type" with "IOA-triggered change re-reporting", only one alarm is generated for a stationary target with the same ID, and re-reporting is only carried out when the target's shape / position changes significantly. This completely solves the problem of repeated alarms in consecutive frames in traditional systems and effectively eliminates meaningless repeated alarms.

[0054] 3. Significantly improved tracking accuracy: The addition of the constraint that "different categories of target boxes belong to different targets" combined with the "multi-IOA matching and selection" logic completely avoids cross-category tracking errors; at the same time, tracking and alarms are only performed on specified vehicle types to reduce tracking interference from non-target types; scenario-specific IOA threshold configuration and target box smoothing further reduce tracking errors caused by pure offset and detection box errors, ensuring the uniqueness and continuity of tracking IDs;

[0055] 4. Significantly reduced invalid alarms: By specifying vehicle type for verification, alarms are only triggered for high-risk engineering vehicles such as excavators and cranes, avoiding invalid alarms caused by non-specified vehicle types such as agricultural vehicles and ordinary trucks; IOA has a higher tolerance for partial obstruction, reducing false alarms caused by trees / towers and significantly lowering the false alarm rate.

[0056] 5. The system has strong real-time performance and is compatible with low computing power devices: IOA calculation does not require the calculation of the union area, which reduces one core operation step compared to IOU. Combined with optimizations such as lightweight target detection model, image downsampling, and simplification of historical boxes, the computing and storage overhead is greatly reduced. It can run in real time on edge computing devices along the power transmission line, and the end-to-end latency can be controlled within 1 second, meeting the real-time requirements of power operation and maintenance.

[0057] 6. High accuracy in change judgment: Combining the combined judgment formula of "same category IOA (spatial position / morphological change) + Pearson coefficient (image feature)", it can accurately identify the changes in the operational behavior of targets of the same category and specified vehicle type, without missing key alarm information and avoiding invalid re-reporting;

[0058] 7. Lightweight and easy-to-deploy system: The historical database only stores core information and scene thresholds, eliminating the need for large-scale storage of historical images and reducing system storage overhead; the parameters of each module and the list of specified vehicle types can be flexibly configured, and it can seamlessly connect to existing power line monitoring cameras and maintenance platforms without the need for large-scale modification of existing equipment, thus reducing project implementation costs;

[0059] 8. Strong degradation operation capability: The system modules are decoupled. Even if the database module fails in a historical scenario, basic monitoring can still be achieved through the core process of "detection-IOA tracing-alarm" to ensure the normal operation of the system's core functions. Attached Figure Description

[0060] Figure 1 A simplified flowchart of the invention process;

[0061] Figure 2 A simplified diagram of the IOA tracing sub-process;

[0062] Figure 3 This is a simplified flowchart of the change determination sub-process. Detailed Implementation

[0063] The embodiments of the present invention will be described in detail below. Although the present invention will be described and illustrated in conjunction with some specific embodiments, it should be noted that the present invention is not limited to these embodiments. On the contrary, any modifications or equivalent substitutions made to the present invention should be covered within the scope of the claims of the present invention.

[0064] Furthermore, to better illustrate the present invention, numerous specific details are set forth in the following detailed embodiments. Those skilled in the art will understand that the present invention can be practiced without these specific details.

[0065] This application proposes a method for suppressing false alarms and eliminating duplicate alarms in the monitoring of power transmission line engineering vehicles. A corresponding system for implementing this method is also designed. The core architecture utilizes lightweight target detection, precise tracking with category-constrained IOA, first-frame reporting of specified types, and change-triggered re-reporting (replacing traditional IOU with IOA, i.e., intersection / current target bounding box area). This approach accurately controls duplicate alarms while reducing false negatives, balancing system real-time performance and scenario adaptability. The specific technical solutions for the method and system are as follows:

[0066] I. Core Method Execution Flow

[0067] The main implementer of this method is the power transmission line engineering vehicle monitoring system, deployed on edge computing devices or cloud servers along the power transmission line. The overall process is as follows: Figure 1 As shown.

[0068] Detailed steps:

[0069] 1. Image acquisition module (including location number): Cameras at each monitoring point of the transmission line continuously acquire real-time images within the protected area, add a unique location number and timestamp to each frame of the image, and transmit the tagged image data to the detection module. The location number is used to associate the location with the historical scene database.

[0070] 2. Lightweight Engineering Vehicle Detection Module: This module employs lightweight deep learning object detection models (such as YOLOv5n and YOLOv8n) to infer the characteristics of labeled real-time images, identifying engineering vehicle targets within the images and outputting the bounding box coordinates (x1, y1, x2, y2), target category confidence score, and vehicle category. Valid bounding boxes are then filtered using a confidence score formula, as follows:

[0071]

[0072] in: The confidence level (0-1) of the engineering vehicle category output by the target detection model. To pre-set the confidence threshold (0.7 in this scheme), only target boxes with a confidence level ≥ 0.7 are retained to obtain a list of valid detected targets.

[0073] 3. IOA Tracking Module (Replacing Traditional IOU): Initializes the tracking list to maintain valid tracked targets in consecutive frames. Each target item in the tracking list includes a tracking ID, target category, latest bounding box, and count of undetected frames. For valid detected target boxes in the current frame, the IOA (intersection area / current target box area) is calculated one by one with the target boxes in the tracking list of the previous frame, taking into account category constraints. The IOA calculation formula is as follows:

[0074]

[0075] Specific calculation: Set the current target bounding box A: History frame B: ,but:

[0076] Coordinates of the top left corner of the intersection region: , ;

[0077] Coordinates of the bottom right corner of the intersection region: , ;

[0078] Intersection area: ;

[0079] Area of ​​target box A:

[0080] IOA tracking sub-processes are as follows Figure 2 As shown.

[0081] Double constraints: Target boxes of different categories belong to different targets, and IOA matching calculation is only performed on target boxes of the same category and belonging to the specified vehicle type, completely avoiding mismatches across categories or non-specified types;

[0082] IOA matching rules:

[0083] If the IOA value of the current target box and a target box of the same category in the previous frame is greater than the threshold (default 0.5), they are determined to be the same target. The tracking ID and the latest bounding box coordinates of the target are updated, and the number of undetected frames of the ID is reset to 0.

[0084] Specifically: If the IOA of the current target box and multiple target boxes of the same category in the previous frame are all greater than the threshold, then the target box with the largest IOA value is selected for matching, and the other matching items are considered invalid, to ensure the uniqueness of the tracking ID;

[0085] If the IOA value of the current target box and all target boxes of the same category in the previous frame is less than or equal to the threshold, it is determined to be a new target, a unique new tracking ID is assigned to it and it is added to the tracking list, and the number of undetected frames is initialized to 0; if the IOA value of the current target box and a target box in the previous frame is greater than 0.5, it is determined to be the same target, and the tracking ID and the latest bounding box coordinates of that target are updated.

[0086] Scenario-specific threshold adaptation: IOA thresholds support flexible configuration based on scenarios—0.4 for points around construction sites (adapting to shape changes), 0.6 for points in urban areas (adapting to pure offset), and 0.5 for points in suburban areas (general).

[0087] ID cleanup rules: For IDs that are not matched in the tracking list, increment the number of frames that have not been detected by 1; if the target corresponding to a tracking ID is not detected for 3 consecutive frames, remove it from the tracking list, and finally output a list of targets with unique tracking IDs.

[0088] 4. Alarm Management Module (Alarm Generation and Dynamic Re-reporting Rules): Includes a built-in ID alarm record dictionary and change judgment unit. The key of the ID alarm record dictionary is the tracking ID, and the values ​​are (first frame reporting status, target category, historical target box list (only the most recent 3 frames are retained), previous frame target box, previous frame region image). The core rule is "specified type first frame reporting + IOA trigger change re-reporting," and it achieves bidirectional interaction with the historical scene database.

[0089] First frame reporting: When a new ID appears for the first time, if the target type belongs to the specified vehicle type, an alarm is immediately generated and marked "first frame reported". The alarm information includes the location number, timestamp, tracking ID, target category, and target bounding box. At the same time, the initial target bounding box and the corresponding area image of the ID are recorded. If the target type does not belong to the specified vehicle type, it is only added to the tracking list but the first frame reporting is not triggered.

[0090] Change-triggered re-reporting: For already reported IDs (and belonging to the specified vehicle type), first, retrieve the historical target bounding box information of that location from the historical scene database based on the location number, then perform target change determination. The change determination sub-process is as follows: Figure 3 As shown.

[0091] The core formula for change determination (combination condition, which must be satisfied simultaneously) is: the IOA value of the current frame's bounding box and the k-th frame's bounding box of the same category in the past 3 frames.

[0092]

[0093] in:

[0094] The IOA value is the sum of the current frame bounding box and the k-th frame bounding box of the same category from the previous 3 frames. IOA thresholds are configured for different scenarios (0.4 for construction sites, 0.6 for urban areas, and 0.5 for suburban areas).

[0095] The Pearson correlation coefficient is the sum of the current frame's and the previous frame's target regions. The formula for calculating the Pearson coefficient is as follows:

[0096]

[0097] A flattened array of grayscale values ​​of the target region in the current frame after downsampling. This is a flattened array of grayscale values ​​from the same target region with the same ID in the previous frame. , They are respectively , The mean, The value range is -1 to 1, and this scheme takes a non-negative value (0 to 1).

[0098] Re-reporting rules: When the combined conditions for change judgment are met, and the target is determined to have changed significantly (such as crane boom extension, excavator boom rotation), a re-reporting alarm is triggered, and the alarm information is updated with the latest target position and status; at the same time, the historical target box record for that ID is deleted, and only the current box is retained as the new "historical baseline". If the conditions are not met, only the target box / region image is updated, and no alarm is generated;

[0099] Error optimization: Add target box smoothing processing (take the average coordinates of the target boxes in the current frame and the previous frame) before IOA calculation to offset the impact of random deviation of the detection box on IOA calculation;

[0100] 5. Dynamic updates to the historical scene database: The alarm management module associates the valid targets detected this time (including tracking ID, target category, latest bounding box, and scene IOA threshold) with the location number and updates them to the historical scene database of that location, realizing lightweight dynamic management of the scene.

[0101] II. System Architecture for Implementing the Above Methods

[0102] This application also proposes a power transmission line engineering vehicle monitoring system that implements the above-mentioned methods for suppressing false alarms and eliminating repetitive alarms. This system has a modular architecture and is deployed on a power transmission line operation and maintenance monitoring platform. The modules work collaboratively, specifically including:

[0103] 1. Image acquisition module: Composed of high-definition cameras at each monitoring point in the power transmission line protection zone and a data transmission unit, it is used to acquire real-time images, add point numbers and timestamps to the images, and transmit the image data to subsequent modules;

[0104] 2. Lightweight Engineering Vehicle Detection Module: It has a built-in pre-trained lightweight engineering vehicle target detection model and a list of specified vehicle types, which is used to identify and classify engineering vehicles in real-time images. It filters low-confidence detection results through a confidence filtering formula and outputs a list of valid detection targets that contain category information and belong to the specified vehicle type.

[0105] 3. IOA Tracking Module: It has built-in IOA calculation algorithm, category matching logic, scene-specific threshold configuration module and tracking ID management logic. It only performs IOA matching on target boxes of the same category and belonging to the specified vehicle type, handles the selection logic of multiple IOA matching, and executes the ID cleaning rule for three consecutive frames that have not been detected. It outputs a list of targets with unique tracking IDs.

[0106] 4. Alarm Management Module: Includes a built-in ID alarm record dictionary, specified vehicle type verification logic, target bounding box smoothing unit, and change determination unit. The change determination unit performs image downsampling, calculation of the maximum IOA value for the same category, and Pearson coefficient calculation. It only applies the alarm rule of "first frame reporting + IOA triggering change re-reporting" to specified vehicle types to eliminate duplicate alarms. Simultaneously, it integrates with the historical scene database to acquire and update data.

[0107] 5. Historical Scene Database Module: Establishes a lightweight historical scene database for each monitoring point, storing only the point number, tracking ID, target category, latest target bounding box information, and scene IOA threshold, supporting fast database retrieval and overwrite updates;

[0108] 6. Main Control Module: As the core scheduling unit of the system, it connects all the above modules, controls the execution order, data transmission and interaction of each module, and supports parameter configuration of each module (such as IOA scene threshold, Pearson coefficient threshold, downsampling size, ID cleanup frame number threshold, and specified vehicle type list).

[0109] (III) Configuration of Key System Parameters

[0110] The core parameters of this system can be flexibly configured according to the actual scenario of each monitoring point on the transmission line. The default optimal parameters are as follows:

[0111]

[0112] The default list of specified vehicle types includes: cranes, concrete pump trucks, excavators, pile drivers, and bulldozers. This invention also has the following features:

[0113] 1. Algorithm Adaptability: The lightweight object detection model (YOLOv5n / YOLOv8n) used in this method is an open-source pre-trained model that can be fine-tuned using a power transmission line engineering vehicle dataset (including specified types such as excavators, cranes, and pile drivers) to further improve the accuracy of engineering vehicle recognition and classification. IOA calculation and Pearson coefficient calculation are both fundamental algorithms in the field of computer vision, which are simple to implement, computationally efficient, and do not require complex model training and deployment.

[0114] 2. Hardware Deployment Requirements: The edge deployment hardware requirements for this system are: CPU ≥ 4 cores and memory ≥ 8G, which can meet the conventional hardware deployment conditions of monitoring points along the power transmission line; the cloud deployment can directly connect to the existing power operation and maintenance cloud platform to realize unified management of data from multiple points and alarm push.

[0115] 3. Threshold adaptation suggestions:

[0116] a. Urban locations (high interference, rapid target changes): The Pearson coefficient threshold can be set to 0.6, and the IOA threshold can be set to 0.6 to improve the sensitivity of change detection; "Aerial work vehicle" can be added to the list of specified vehicle types;

[0117] b. Suburban locations (less interference, slower target changes): The Pearson coefficient threshold can be set to 0.8, and the IOA threshold can be set to 0.5 to reduce invalid re-reporting;

[0118] c. Locations around the construction site (multiple types of vehicles gather, and their shapes change frequently): The IOA threshold is set to 0.4 to adapt to scenarios where the target moves frequently and changes shape, while strengthening category constraints to avoid cross-category mismatches.

[0119] 4. Actual test results: This method and system have completed phased tests at various monitoring points in the power transmission line protection zone. The test results show that compared with the existing IOU-based monitoring system, the false alarm rate is reduced by more than 40%, the duplicate alarm problem is reduced by more than 90%, the cross-category tracking error rate and the non-specified type invalid alarm rate are significantly reduced, and the system end-to-end latency is controlled within 1 second, which can meet the actual application needs of power operation and maintenance.

Claims

1. A method for suppressing false alarms and eliminating repetitive alarms in the monitoring of power transmission line engineering vehicles, characterized in that, Includes the following steps: Step S1: Collect real-time images by using cameras deployed at various monitoring points in the power transmission line protection zone, and add a unique location number and timestamp to each frame of the image; Step S2: Use a lightweight target detection model to identify engineering vehicles in the real-time image, output the bounding box coordinates, target category and confidence score of the target, and filter the valid target boxes through the confidence score filtering formula to obtain a list of valid detected targets; Step S3: Based on the IOA algorithm, perform cross-frame tracking and ID assignment on valid detected target boxes in consecutive frames, where IOA is the ratio of the intersection area to the current target box area, and output a list of targets with unique tracking IDs; Step S4: Execute alarm generation and duplicate alarm elimination rules, including: when a new tracking ID appears for the first time, if the target type belongs to the specified vehicle type, it should be reported immediately in the first frame; for a track ID that has already been reported, when the change judgment condition is met, a re-report alarm is triggered and the historical target box record of the tracking ID is deleted. Step S5: Associate the detected valid target information with the location number and update it to the historical scene database.

2. The method according to claim 1, characterized in that, In step S2, a list of specified vehicle types is preset, and valid target boxes are filtered using a confidence-based filtering formula: in, The confidence score for the engineering vehicle category output by the target detection model, with a value ranging from 0 to 1; A preset confidence threshold of 0.7 is set; only target boxes with a confidence level greater than or equal to the preset confidence threshold are retained, and a list of valid detected targets containing target category information and belonging to the specified vehicle type is output.

3. The method according to claim 2, characterized in that, Step S3, which involves cross-frame tracking and ID allocation based on the IOA algorithm, includes: Constraints: Target boxes of different categories belong to different targets; IOA matching is only performed on target boxes of the same category and belonging to the specified vehicle type. The formula for calculating IOA is: in, The target bounding box for the current frame. This is the historical target bounding box from the previous frame. For the target box With target box The area of ​​intersection For the target box The area; IOA matching rules: If the IOA value of the current target box and a target box of the same category in the previous frame is greater than a preset threshold, they are determined to be the same target. The tracking ID and the latest bounding box coordinates of the target are updated, and the number of undetected frames of the tracking ID is reset to zero. If the IOA of the current target box and multiple target boxes of the same category in the previous frame are all greater than the preset threshold, then the target box with the largest IOA value is selected for matching. If the IOA value of the current target box and all target boxes of the same category in the previous frame are less than or equal to the preset threshold, it is determined to be a new target, a unique new tracking ID is assigned to it and it is added to the tracking list; ID cleanup rules: For tracking IDs that are not matched in the tracking list, increment the number of frames not detected by one; if the target corresponding to a tracking ID is not detected for three consecutive frames, remove it from the tracking list.

4. The method according to claim 3, characterized in that, The preset threshold is configured according to the scene type of the monitoring point: 0.4 for points around the construction site, 0.6 for points in urban areas, and 0.5 for points in suburban areas; and the target box is smoothed before IOA calculation, taking the average coordinates of the target box in the current frame and the previous frame.

5. The method according to claim 4, characterized in that, The change determination condition in step S4 is that the following two sub-conditions must be met simultaneously: in, k∈{1,2,3} represents the bounding box of the current frame and the bounding box of the same category in the three historical frames. The IOA value of the frame target bounding box. IOA thresholds configured for different scenarios; The Pearson correlation coefficient between the target regions of the current frame and the previous frame. This is the threshold for the Pearson coefficient.

6. The method according to claim 5, characterized in that, The formula for calculating the Pearson correlation coefficient is as follows: in, A flattened array of grayscale values ​​of the target region in the current frame after downsampling. This is a flattened array of grayscale values ​​from the same tracking ID target region in the previous frame after downsampling. , They are respectively , The average value; the Pearson coefficient threshold is configured according to the scenario: 0.6 for urban points, 0.8 for suburban points, and 0.7 for points around construction sites.

7. The method according to any one of claims 1-6, characterized in that, In step S4, before determining the target change, the region image is downsampled to 64×64 pixels, and only the most recent three frames are retained in the historical target box; the specified vehicle type list by default includes cranes, pump trucks, excavators, pile drivers, and bulldozers.

8. A power transmission line engineering vehicle monitoring system implementing the method of any one of claims 1-6, characterized in that, include: The image acquisition module is used to acquire real-time images through high-definition cameras at various monitoring points in the power transmission line protection zone, and add point numbers and timestamps to the images; The lightweight engineering vehicle detection module has a built-in pre-trained lightweight engineering vehicle target detection model and a list of specified vehicle types. It is used to identify and classify engineering vehicles in real-time images, and outputs a list of valid detection targets that contain category information and belong to the specified vehicle type through a confidence filtering formula. The IOA tracking module has built-in IOA calculation algorithm, category matching logic, scene-specific threshold configuration module and tracking ID management logic. It only performs IOA matching on target boxes of the same category and belonging to the specified vehicle type, handles the selection logic of multiple IOA matching, and executes ID cleanup rules for three consecutive frames that have not been detected. The alarm management module has a built-in ID alarm record dictionary, specified vehicle type verification logic, target box smoothing processing unit and change judgment unit. It only executes the alarm rule of first frame reporting plus IOA triggering change re-reporting for specified vehicle types. The historical scene database module establishes a lightweight historical scene database for each monitoring point, storing the point number, tracking ID, target category, latest target bounding box information, and scene IOA threshold. The main control module connects all the above modules and controls the execution order, data transmission, and interaction of each module.

9. The system according to claim 8, characterized in that, The historical scene database module retains three frames of historical target boxes; the list of specified vehicle types can be flexibly configured through the main control module.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method described in any one of claims 1-6.

Images