Container abnormal state detection method and system based on dynamic patrol

Abstract

The application discloses a box abnormal state detection method and system based on dynamic patrol. The method comprises the following steps: acquiring patrol video and vehicle pose information; performing box target detection and cross-frame tracking on the video to obtain a continuous observation sequence; performing spatial intersection on observation lines of sight of the same box target, and combining vehicle pose information to determine the absolute spatial position of the box target; performing abnormal detection on key components of the box and extracting abnormal geometric features; correlating multi-period inspection data, calculating abnormal evolution features, and generating a state code; predicting a health trend, calculating a health index and a risk level after combining with environmental factors, and generating an early warning and operation and maintenance task, and updating a box health record. The method improves the accuracy of abnormal detection and the inspection efficiency under the dynamic patrol environment, and provides reliable data support for intelligent operation and maintenance of facilities.

Description

Technical Field

[0001] This invention relates to computer vision, and more specifically to a method and system for detecting abnormal states of enclosures based on dynamic inspection. Background Technology

[0002] With the continuous advancement of urban infrastructure construction, various box-type facilities such as power junction boxes, communication cabinets, and gas boxes are widely distributed in public areas such as roads, parks, and residential areas, undertaking key functions such as power transmission, communication support, and municipal facility management. Currently, box inspection mainly adopts three methods: manual inspection, fixed monitoring, and intelligent detection based on image recognition. Manual inspection relies on inspectors to check point by point, which has inherent drawbacks such as low inspection efficiency, limited coverage, and high maintenance costs; fixed monitoring is strictly limited by installation location and monitoring range, making it difficult to meet the actual needs of large-scale dynamic road inspection; existing image recognition-based detection solutions mostly target static images for abnormal state identification. Although they can improve detection efficiency to some extent, their analysis dimension is limited to instantaneous anomalies in the current image, lacking the ability to comprehensively judge the historical state and changing trends of the target.

[0003] Because inspection vehicles are constantly moving, the acquired images are easily affected by factors such as vehicle body vibration, changes in lighting, motion blur, and occlusion. Furthermore, the shooting angle and spatial position of the same enclosure vary significantly across different inspection cycles, leading to problems such as false positives, false negatives, and duplicate identifications in existing single-frame image-based detection methods, making it difficult to guarantee the positioning accuracy of the enclosure targets. In addition, existing technologies for identifying enclosure anomalies typically remain at the level of category determination, lacking refined quantitative analysis of abnormal features such as rust, cracks, tilting, and door opening, making it difficult to objectively reflect the severity of the anomalies. More critically, existing detection systems lack the ability to fuse and analyze multi-source data including vehicle positioning information, historical inspection data, and environmental factors. This makes it impossible to establish a continuous tracking mechanism for enclosure anomalies, distinguish between short-term, occasional anomalies and continuously deteriorating anomalies, and even more difficult to predict anomaly development trends and assess remaining lifespan. Simultaneously, existing inspection systems typically only store single inspection results, lacking historical data accumulation and status update mechanisms, making it difficult to create a full life-cycle health record for the enclosure and failing to provide continuous data support for subsequent inspections and maintenance decisions.

[0004] In summary, existing technologies in dynamic inspection environments suffer from several core technical problems, including low accuracy in container target positioning, insufficient refinement of abnormal feature extraction, lack of continuous tracking and evolution analysis capabilities for abnormal states, lack of risk assessment and intelligent early warning capabilities based on multi-factor fusion, and lack of full lifecycle management of inspection data.

[0005] Therefore, it is necessary to design a new method to improve the accuracy and efficiency of anomaly detection in dynamic inspection environments, and to provide reliable data support for intelligent operation and maintenance of facilities. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for detecting abnormal states of enclosures based on dynamic inspection.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for detecting abnormal states of enclosures based on dynamic inspection, comprising: Acquire road environment videos collected by the patrol vehicle during its operation, as well as vehicle pose information corresponding to each video frame; Box target detection and cross-frame tracking are performed on the road environment video to obtain a continuous observation sequence; Based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, the observation lines of the same box target in different video frames in the continuous observation sequence are converted to the same reference coordinate system and then spatial line intersection is performed. The relative spatial position obtained by the intersection is converted into an absolute spatial position by combining the vehicle pose information. Anomaly detection is performed on the key component areas of the box target, and abnormal geometric features are extracted; Based on the absolute spatial location, the abnormal geometric features of the same container target under different inspection cycles are associated and matched, and a container identifier is assigned to the same container target. The abnormal evolution features are calculated and an abnormal evolution status code is generated. Based on the abnormal evolution characteristics, the abnormal evolution status codes, and historical inspection data, the health status change trend of the container target is predicted, and corrections are made in conjunction with environmental influencing factors. The container health index and comprehensive risk value are calculated, and the risk level is determined. Based on the risk level, early warning information and operation and maintenance task information are generated, and the health record of the container target is updated.

[0008] The further technical solution is as follows: Based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, the observation lines of the same box target in different video frames in the continuous observation sequence are converted to the same reference coordinate system, and spatial line intersection is performed. The relative spatial position obtained from the intersection is then converted to an absolute spatial position by combining the vehicle pose information, including: A camera projection model is established based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, and the observation line of view corresponding to the feature points of the box target in different video frames is calculated. Intersection calculations are performed on the observation lines of the same feature point under different video frame combinations to obtain multiple relative spatial positions; The relative spatial positions of each video frame are fused and optimized based on the target detection confidence and spatial calculation error. Based on the installation position parameters of the vehicle-mounted camera relative to the patrol vehicle and the vehicle's pose information, the fused and optimized relative spatial position is converted to a geographic coordinate system to obtain the absolute spatial position.

[0009] The further technical solution is as follows: the anomalies include planar anomalies, linear anomalies and state anomalies. The planar anomalies include rust, paint peeling, corrosion and dirt. The linear anomalies include cracks and fissures. The state anomalies include open cabinet doors, tilted cabinets and missing parts. The extraction of abnormal geometric features includes: For the somatic anomalies, the abnormal region is separated from the normal region, the boundary contour of the abnormal region is extracted, and the area and proportion of the abnormal region are calculated. For the linear anomaly, the skeleton of the anomaly region is extracted, the skeleton coordinates are extracted, and the crack length, crack width and crack direction are calculated. For the aforementioned status anomalies, the door opening angle is calculated based on the positional relationship between the door area and the main body of the box, the box tilt angle is calculated based on the edge contour of the box and the vertical reference direction, and the missing component is determined based on the integrity detection of the key component area.

[0010] The further technical solution is as follows: the association and matching of the abnormal geometric features of the same container target under different inspection cycles based on the absolute spatial position includes: Based on the absolute spatial location and historical inspection data, determine whether the container targets in different inspection cycles belong to the same container target; For the planar anomalies, matching is performed based on the boundary contour of the anomaly region; for the linear anomalies, matching is performed based on the skeleton coordinates and the crack direction; for the state-related anomalies, matching is performed based on the door opening angle and the box body tilt angle; for the component missing anomalies, matching is performed based on the key component category, the missing part, and the component integrity detection results. The abnormal regions extracted in each inspection cycle are projected onto a geometric reference plane and normalized.

[0011] Its further technical solution is: the calculation of abnormal evolution characteristics and the generation of abnormal evolution status codes include: The abnormal evolution characteristics are calculated based on the changes in the abnormal geometric features during different inspection weeks. The abnormal evolution characteristics include the abnormal area growth rate, the abnormal boundary propagation rate, the crack length growth rate, the crack width change rate, the crack direction change, the box attitude change rate, the abnormal duration, and the number of abnormal recurrences. The abnormal development stages are classified according to the abnormal evolution characteristics, and the abnormal evolution status codes are generated. The abnormal evolution status codes include the first status code to the fifth status code in ascending order of the severity of the development stages. The abnormal evolution features and the abnormal evolution status codes are used to construct an abnormal evolution feature vector.

[0012] The further technical solution is as follows: predicting the health status change trend of the container target based on the abnormal evolution characteristics, the abnormal evolution status code, and historical inspection data includes: Using the abnormal evolution feature vector as input and combining it with the abnormal change patterns in historical inspection data, the health status change trend of the container target in future inspection cycles is predicted to obtain the prediction result. The abnormal evolution features and the prediction results in the abnormal evolution feature vector are weighted and fused to obtain the box health index.

[0013] The further technical solution is as follows: the modification based on environmental impact factors includes: The environmental impact factors include road grade, traffic flow, rainfall, air humidity, salt spray environment, industrial pollution, and obstruction. When the container target is in a high humidity environment or salt spray environment, the influence weight of the planar anomalies corresponding to rust and corrosion is increased; when the traffic flow in the area where the container target is located is greater than a preset traffic flow threshold, the influence weight of the linear anomalies and the state-type anomalies is increased. The container health index is updated based on the corrected impact weights of each anomaly, and the overall risk value is updated again using the updated container health index to obtain the environmental correction result.

[0014] Its further technical solution is as follows: the calculation of the health index and comprehensive risk value of the enclosure, and the determination of the risk level, includes: The comprehensive risk value is calculated based on the health index of the enclosure, the prediction result, and the environmental correction result, and the risk level is divided into normal, low risk, medium risk, high risk, and severe risk according to the preset risk threshold. The prediction confidence level is calculated based on historical prediction error, anomaly detection confidence level, and historical data completeness. The prediction confidence level is not included in the calculation of the comprehensive risk value. When the prediction confidence level is lower than a preset confidence level threshold, the corresponding prediction result is marked to indicate the need for review.

[0015] The further technical solution is as follows: generating early warning information and operation and maintenance task information based on the risk level, and updating the container health record of the container target, includes: When the risk level is normal, record the current health status; when the risk level is low or medium risk, generate a reminder or inspection and review task; when the risk level is high or severe risk, generate a key handling warning. Generate operation and maintenance task information, which includes the container identifier, the absolute spatial location, the anomaly type, the risk level, the suggested handling measures, and the task priority; Obtain maintenance feedback information including maintenance completion status and anomaly elimination status; update the health index of the enclosure when the anomaly is eliminated after maintenance; retain historical anomaly records and increase the attention level of subsequent inspections when the anomaly still exists. New inspection data, abnormal changes, and maintenance feedback information are written into the health record of the enclosure.

[0016] This invention also provides a container abnormality detection system based on dynamic inspection, comprising: The data acquisition unit is used to acquire road environment videos collected by the patrol vehicle during its operation, as well as vehicle pose information corresponding to each video frame. The target tracking unit is used to perform box target detection and cross-frame tracking on the road environment video to obtain a continuous observation sequence; The spatial positioning unit is used to perform spatial line intersection on the observation lines of the same box target in different video frames in the continuous observation sequence, based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, and the relative spatial position obtained by the intersection is converted to the same reference coordinate system after the observation lines of the same box target in different video frames are converted to the absolute spatial position by combining the vehicle pose information. An anomaly detection unit is used to detect anomalies in key component areas of the box target and extract abnormal geometric features; The evolution analysis unit is used to perform correlation matching of the abnormal geometric features of the same box target under different inspection cycles based on the absolute spatial position, assign box identifiers to the same box target, calculate abnormal evolution features, and generate abnormal evolution status codes. The risk assessment unit is used to predict the health status change trend of the container target based on the abnormal evolution characteristics, the abnormal evolution status code and historical inspection data, make corrections in combination with environmental impact factors, calculate the container health index and comprehensive risk value, and determine the risk level. The early warning closed-loop unit is used to generate early warning information and operation and maintenance task information according to the risk level, and update the container health record of the container target.

[0017] The beneficial effects of this invention compared to existing technologies are as follows: This invention achieves precise positioning of the enclosure by integrating vehicle-mounted video and positioning information to perform spatial line-of-sight intersection, extracts refined geometric features of key component anomalies, calculates anomaly evolution features based on absolute spatial position and multi-cycle data and generates status codes, and predicts health trends and assesses risk levels by combining historical data and environmental factors. Based on this, it generates intelligent early warning and maintenance tasks and dynamically updates health records, thereby solving problems such as inaccurate positioning, insufficient feature quantification, lack of trend tracking and fragmented data management in dynamic inspections. This significantly improves the accuracy of anomaly detection and inspection efficiency, and provides full lifecycle data support for intelligent facility operation and maintenance.

[0018] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating the abnormal state detection method for enclosures based on dynamic inspection provided in this embodiment of the invention; Figure 2 This is a schematic diagram of multi-box detection and cross-frame correlation in continuous video frames provided in an embodiment of the present invention. Figure 1 ; Figure 3 This is a schematic diagram of multi-box detection and cross-frame correlation in continuous video frames provided in an embodiment of the present invention. Figure 2 ; Figure 4 This is a schematic diagram of multi-box detection and cross-frame correlation in continuous video frames provided in an embodiment of the present invention. Figure 3 ; Figure 5 This is a schematic diagram illustrating the location of key components of the enclosure and the quantification of multiple types of abnormal features provided in an embodiment of the present invention. Figure 6 This is a schematic diagram of box tilt anomaly detection and tilt angle calculation provided in an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the calculation of health index, trend prediction, and comprehensive risk assessment provided in an embodiment of the present invention. Figure 8 A schematic block diagram of a container abnormality detection system based on dynamic inspection provided in an embodiment of the present invention; Figure 9 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] Please see Figure 1 , Figure 1This is a flowchart illustrating the container anomaly detection method based on dynamic inspection provided in this embodiment of the invention. This method, applied to a server, achieves precise container target location by fusing vehicle-mounted video and GNSS positioning information for spatial line-of-sight intersection. It performs refined geometric feature extraction and quantitative analysis of anomalies such as corrosion, cracks, and door opening. Based on absolute spatial location and multi-cycle inspection data, it calculates anomaly evolution characteristics and generates status codes. Combining historical inspection data with environmental factors, it predicts health trends, calculates health indices and comprehensive risk values ​​to determine risk levels. Based on this, it generates tiered early warnings and maintenance tasks and dynamically updates the container health record. This effectively solves problems such as low target positioning accuracy, coarse anomaly feature extraction, lack of continuous tracking and evolution analysis, single risk assessment, and fragmented data management in dynamic inspections. It significantly improves the accuracy of anomaly detection and inspection efficiency, providing continuous and reliable data support for intelligent facility operation and maintenance.

[0026] Figure 1 This is a flowchart illustrating the abnormal state detection method for enclosures based on dynamic inspection provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S110 to S170.

[0027] S110. Acquire road environment videos collected by the patrol vehicle during its operation, as well as vehicle pose information corresponding to each video frame.

[0028] In this embodiment, the road environment video refers to video data obtained by continuously capturing the roadside environment through a vehicle-mounted camera fixedly installed on the patrol vehicle during the vehicle's operation. The road environment video includes image information of roadside box facilities (such as power distribution boxes, communication junction boxes, traffic signal control boxes, street light control boxes, etc.). The vehicle-mounted camera is only used to collect visual information about the road environment, including visible light cameras, infrared cameras, or other imaging devices capable of acquiring road environment images.

[0029] Vehicle pose information refers to the position and attitude data of the patrol vehicle corresponding to the acquisition time of each video frame. The position data includes longitude, latitude, elevation, and corresponding timestamps, while the attitude data includes heading angle, pitch angle, and roll angle. The position data is acquired through a Global Navigation Satellite System (GNSS) receiver or other positioning equipment, and the attitude data is obtained through GNSS dual-antenna orientation, inertial measurement unit (IMU), or vehicle wheel speed and steering information calculation. In other embodiments, the vehicle pose information can also be obtained through RTK-GNSS high-precision positioning, GNSS / INS integrated navigation, UWB positioning, or visual positioning based on high-precision map matching.

[0030] Specifically, during the patrol vehicle's operation, the onboard camera continuously captures road environment video at a preset frame rate. The system simultaneously records the capture time information of each video frame and the vehicle's pose information corresponding to that capture time. It also establishes a spatiotemporal correlation between the video data and the vehicle's movement process through timestamp alignment, ensuring that each video frame has a corresponding vehicle pose reference. In other embodiments, the patrol vehicle can also be replaced by a patrol robot, drone, or other mobile patrol platform.

[0031] By fusing vehicle video with positioning information through the above data acquisition method, a unified time and space reference is provided for the subsequent spatial location calculation of the target box. Compared with the detection method that only relies on a single frame image, it can effectively solve the problems of repeated identification of the same box, target confusion and large positioning error when the vehicle is in motion.

[0032] S120. Perform box target detection and cross-frame tracking on the road environment video to obtain a continuous observation sequence.

[0033] In this embodiment, a continuous observation sequence refers to a set of observation data arranged in chronological order, formed by associating the detection results of the same container target in multiple consecutive video frames. Each observation record in the continuous observation sequence includes at least the frame identifier of the video frame, the position coordinates of the container target in the image, the bounding box information, and the detection confidence. The detection confidence is a value output by the target detection model that characterizes the degree of confidence that the detected container target is a real container facility, and its value ranges from 0 to 1.

[0034] Please see Figures 2 to 4 The video frames of the road environment video are preprocessed, including image denoising, brightness adjustment, distortion correction, and image enhancement, to reduce the impact of vehicle movement, changes in ambient lighting, and changes in shooting angle on subsequent target recognition and spatial calculation processes. Image denoising can employ Gaussian filtering, median filtering, or a deep learning-based denoising network; brightness adjustment can use histogram equalization or gamma correction; distortion correction is performed based on the lens distortion parameters of the vehicle-mounted camera; and image enhancement can use contrast stretching or sharpening.

[0035] Box-shaped targets are detected on the preprocessed video frames. The target detection can employ a deep learning-based target detection model, traditional image detection methods, or other methods capable of target localization. The system learns the appearance, structural, and spatial features of the box-shaped targets to locate the box-shaped regions in complex road environments and outputs the position coordinates, bounding box information, and detection confidence of the box-shaped targets.

[0036] Cross-frame tracking of box targets in continuous video frames: By extracting the appearance features, edge features, corner features, and texture features of the box targets, the system performs association matching on box targets in different video frames to determine whether targets detected in multiple consecutive video frames belong to the same box. Detections belonging to the same box are then associated in chronological order to establish a continuous observation sequence of the box targets. In other embodiments, the cross-frame tracking can also employ motion tracking based on optical flow, state tracking based on Kalman filtering or particle filtering, or target matching methods based on Siamese networks. To address issues such as changes in lighting, shooting angle, and background interference during vehicle inspections, the system can further combine image enhancement, feature fusion, and target tracking algorithms to stably identify box targets in continuous video frames.

[0037] By establishing a continuous observation sequence through detection and cross-frame tracking, the discrete detection results of a single frame are organized into continuous multi-view observation data, avoiding repeated identification and target confusion of the same box body during dynamic inspection. This reduces false detections and missed detections caused by vehicle body bumps, motion blur, and changes in lighting, and improves the accuracy and reliability of box body detection in dynamic inspection environments.

[0038] S130. Based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, the observation lines of the same box target in different video frames in the continuous observation sequence are converted to the same reference coordinate system and then spatial line intersection is performed. The relative spatial position obtained by the intersection is converted into an absolute spatial position by combining the vehicle pose information.

[0039] In this embodiment, relative spatial position refers to the three-dimensional spatial coordinates of the container target in the camera coordinate system, that is, the position coordinates in the coordinate system established with the optical center of the vehicle camera as the origin and the optical axis direction as the reference; absolute spatial position refers to the spatial coordinates of the container target in the geographic coordinate system, including longitude, latitude and elevation, which is used to uniquely identify the geographical location of the container facility in the actual road environment.

[0040] Since a single frame image can only provide projection information of the box target in a two-dimensional image, its positioning result is easily affected by the shooting angle, vehicle vibration and detection error, making it difficult to directly obtain a stable spatial position; therefore, this embodiment utilizes the perspective change relationship generated by continuous video frames during vehicle movement to establish a spatial line intersection model based on temporal multi-view geometric constraints, and realizes the inverse calculation of the spatial position of the box target through steps S131~S135.

[0041] In one embodiment, step S130 described above may include steps S131 to S135.

[0042] S131. Establish a camera projection model based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, and calculate the observation line of view corresponding to the feature points of the box target in different video frames.

[0043] In this embodiment, the camera projection model refers to a mathematical model (such as a pinhole camera model) that describes the mapping relationship between three-dimensional spatial points and two-dimensional image pixels. The internal imaging parameters include the focal length, principal point coordinates, and lens distortion parameters of the vehicle-mounted camera. These internal imaging parameters can be obtained through pre-calibration. Feature points refer to image points with stable and identifiable characteristics extracted from the target image of the box, including edge points, corner points, or center feature points of the box. The observation line of sight refers to the spatial ray that originates from the optical center of the vehicle-mounted camera, passes through the feature points, and corresponds to the pixel points on the imaging plane. The direction of the observation line of sight is calculated from the pixel coordinates of the feature points according to the camera projection model. The process of deriving the direction of the spatial line of sight from the image pixel coordinates is called perspective projection inverse calculation, and its inverse calculation accuracy is determined by the calibration accuracy of the internal imaging parameters.

[0044] S132. Based on the vehicle pose information corresponding to each video frame and the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, the observation line of sight under different video frames is converted to the same reference coordinate system.

[0045] In this embodiment, the system pre-establishes a camera coordinate system, a vehicle coordinate system, and the same reference coordinate system: the camera coordinate system takes the optical center of the vehicle-mounted camera as the origin and the optical axis direction as the forward axis; the vehicle coordinate system takes the phase center of the vehicle positioning antenna as the origin and the vehicle front direction as the forward axis; the same reference coordinate system is a spatial coordinate system that remains fixed during the continuous observation sequence.

[0046] For example, the local station-centered coordinate system determined by the vehicle pose corresponding to the first frame of the continuous observation sequence can be selected, or the East-North-Sky (ENU) horizon coordinate system covering the inspection area can be selected. The installation position parameters (including the position offset of the vehicle-mounted camera relative to the vehicle and the installation height) constitute the translation vector of the origin of the camera coordinate system in the vehicle coordinate system; the installation attitude parameters (including the installation pitch angle, installation yaw angle, and installation roll angle) constitute the rotation matrix from the camera coordinate system to the vehicle coordinate system; the installation position parameters and installation attitude parameters are obtained in advance through calibration after the vehicle-mounted camera is installed, and the two form a fixed homogeneous transformation matrix, that is, the extrinsic parameters of the vehicle-mounted camera relative to the inspection vehicle.

[0047] The vehicle pose information includes vehicle position data (longitude, latitude, and elevation) and vehicle attitude data (heading angle, pitch angle, and roll angle) at the acquisition time of each video frame. The system aligns the timestamps of the video frames with the timestamps of the GNSS positioning data and inertial measurement data. When the output frequency of the positioning data is lower than the video frame rate, linear interpolation is performed on the vehicle position data between two adjacent positioning times, and the vehicle attitude data is recursively calculated based on the angular velocity and acceleration output of the inertial measurement unit. This yields the vehicle pose corresponding to the precise acquisition time of each video frame. Through the above time synchronization processing, it is ensured that the line of sight of each video frame strictly corresponds to the vehicle pose at the actual acquisition time of that frame, avoiding intersection errors introduced by the asynchrony between the pose and the image during vehicle movement.

[0048] For the pixel coordinates of the feature point in the i-th video frame, the system first performs distortion correction on the pixel coordinates based on the lens distortion parameters, and then performs perspective projection inverse calculation based on the intrinsic parameter matrix K of the camera projection model to obtain the observation line direction vector of the feature point in the camera coordinate system. At this time, the observation line is a spatial ray in the camera coordinate system with the optical center as the starting point and the observation line direction vector as the direction. Subsequently, the system transforms the observation line from the camera coordinate system to the same reference coordinate system frame by frame; thus, the observation lines in different video frames are uniformly expressed as spatial rays in the same reference coordinate system, thereby transforming the perspective changes caused by vehicle motion into multi-view observation constraints under the same geometric reference.

[0049] Under the same reference coordinate system, the intersection calculation is performed on multiple observation lines to the same feature point, that is, the spatial point that minimizes the sum of the squares of the distances to each observation line (the optimal intersection point in the least squares sense) is found, which is taken as the relative spatial position of the feature point; after intersecting multiple feature points of the box target respectively, the fusion result is taken, or the observation lines of different video frame combinations are intersected respectively, thus obtaining multiple relative spatial positions.

[0050] S133. Perform intersection calculations on the observation lines of the same feature point after transformation to the same reference coordinate system to obtain multiple relative spatial positions.

[0051] The same feature point under different video frame combinations refers to the set of pixel coordinates of the same feature point in each video frame when the same feature point of the same box target is observed in multiple consecutive video frames; the system establishes geometric constraint relationships between multiple observation lines based on the features of consecutive video frames.

[0052] Specifically, because the position of the onboard camera changes continuously during the inspection vehicle's movement, the same target box will be observed from different perspectives in different video frames. The system utilizes the matching relationships of the same feature points in different video frames, combined with the spatial position changes caused by vehicle movement, to establish geometric constraints between multiple lines of sight. Different combinations of video frames (such as combinations of adjacent frames, interval frames, or multiple frames within a sliding window) are selected to perform spatial line-of-sight intersection calculations, i.e., solving for the optimal intersection point of multiple lines of sight in space (which can be solved using the least squares method), thereby obtaining multiple relative spatial position estimates. The inputs to the spatial line-of-sight intersection calculation include the imaging parameters inside the onboard camera, the positions of the target box feature points in consecutive video frames, the target matching relationships between different video frames, and the position information during vehicle movement. The output is the position coordinates of the target box relative to the camera coordinate system.

[0053] In other embodiments, the spatial positioning can also be achieved by binocular stereo vision parallax ranging, lidar point cloud positioning or structured light ranging, or by fusing RTK-GNSS high-precision positioning results with visual positioning results.

[0054] S134. Based on the target detection confidence and spatial calculation error of each video frame, the multiple relative spatial positions are fused and optimized.

[0055] Specifically, the system first transforms multiple relative spatial positions obtained from different combinations of video frames to the same reference coordinate system based on vehicle motion relationships. Then, considering factors such as target detection confidence, number of feature matches, degree of viewpoint variation, and spatial calculation error for each video frame, it determines the fusion weight of each relative spatial position and performs weighted fusion optimization on multiple relative spatial positions to obtain the optimized relative spatial positions. The higher the detection confidence, the more feature matches, the more sufficient the degree of viewpoint variation, and the smaller the spatial calculation error, the greater the fusion weight of the observation. For example, the fusion weight of each observation can be directly proportional to its target detection confidence and inversely proportional to its spatial calculation error. In other embodiments, the fusion optimization can also employ Kalman filtering, extended Kalman filtering, Bayesian estimation, or graph optimization methods.

[0056] By fusing and optimizing the results of multiple frames of observations, the positional deviation caused by single detection errors, changes in illumination, and vehicle movement was reduced, and a stable spatial position of the box was obtained, providing a reliable foundation for subsequent absolute position calculations.

[0057] S135. Based on the installation position parameters, the installation posture parameters, and the vehicle position information, the fused and optimized relative spatial position is converted to a geographic coordinate system to obtain the absolute spatial position.

[0058] Specifically, the system establishes a transformation relationship between the camera coordinate system, the vehicle coordinate system, and the geographic coordinate system: First, based on the installation position parameters, the fused and optimized relative spatial position is transformed from the camera coordinate system to the vehicle coordinate system. The installation position parameters include the position offset of the vehicle-mounted camera relative to the vehicle, the installation height, and the shooting direction. These installation position parameters are obtained through measurement or calibration after the vehicle-mounted camera is installed. Subsequently, combined with the vehicle pose information, the position in the vehicle coordinate system is transformed to the geographic coordinate system. The absolute spatial position corresponding to the target box is calculated through spatial coordinate transformation, including longitude, latitude, and elevation. The absolute spatial position is then associated with the target box detection results to form basic data of the target box with spatial position identification.

[0059] Through steps S131~S134, spatial line-of-sight intersection is performed by utilizing the perspective change relationship generated by continuous video frames during vehicle movement. Then, through multi-frame fusion and coordinate system transformation, the absolute spatial coordinates of the box target are accurately calculated, which effectively solves the problem of large target positioning error under vehicle movement state and provides a unified spatial benchmark for subsequent anomaly analysis.

[0060] S140. Perform anomaly detection on the key component areas of the box target and extract abnormal geometric features.

[0061] In this embodiment, the critical component area refers to the local area on the enclosure facility that is closely related to functional safety and structural status, including the door area, lock area, enclosure surface area, and base area; abnormal geometric features refer to the feature parameters obtained after pixel-level geometric analysis of the abnormal area, which are used to quantitatively describe the abnormal morphology and degree of damage.

[0062] like Figure 5 As shown, Figure 5 This is a schematic diagram illustrating the location of key components of the enclosure and the quantification of multiple types of abnormal features, provided in an embodiment of the present invention. Figure 5 The left image shows a video frame of the road environment captured by the patrol vehicle during its operation. The system detects box targets in the video frame, obtains the bounding boxes of the box targets, and assigns box identifiers (e.g., ...) to the detected box targets. Figure 5 (ID-201 in the left figure) The box identifier is used to uniquely identify the box facility in order to achieve data association during multiple inspection weeks; Figure 5The right image is a magnified front view of the target enclosure. The system further locates key component areas within the target enclosure area. Label ① indicates the door area, label ② indicates the lock area, label ③ indicates the enclosure surface area, and label ④ indicates the base area. The system extracts corresponding types of abnormal geometric features for different key component areas: for the enclosure surface area, it extracts planar abnormal features, including the area and boundary contour of the abnormal region; for the door area or enclosure surface area, it extracts linear abnormal features, including the skeleton, crack length, and crack width; for the door area and lock area, it extracts security-related abnormal features, including the door opening state and lock state; and for the enclosure and base, it extracts posture-related abnormal features, including the enclosure tilt angle and base state.

[0063] The anomalies include planar anomalies, linear anomalies, and state anomalies. Planar anomalies include rust, paint peeling, corrosion, and dirt. Linear anomalies include cracks and fissures. State anomalies include open doors, tilted enclosures, and missing parts. Specifically, the system can analyze the surface and structural state of the enclosure using methods such as image classification, target detection, semantic segmentation, instance segmentation, and anomaly region recognition to determine whether there are anomalies in the enclosure and to identify the type and location of the anomalies. In other embodiments, other state-related anomalies such as structural deformation may also be detected.

[0064] In one embodiment, the extraction of abnormal geometric features includes steps S141 to S143.

[0065] S141. For the surface anomaly, the anomaly region is separated from the normal region, the boundary contour of the anomaly region is extracted, and the area and proportion of the anomaly region are calculated.

[0066] In this embodiment, the normal region refers to the image region within the target area of ​​the box that has not experienced any anomalies; the boundary contour of the abnormal region refers to the set of pixel coordinates of the dividing line between the abnormal region and the normal region, used to describe the shape and spatial distribution of the abnormal region; the area of ​​the abnormal region refers to the number of pixels contained in the abnormal region or the actual area after scale conversion, used to describe the range of the anomaly's influence; the region proportion refers to the ratio of the area of ​​the abnormal region to the area of ​​the target area of ​​the box (or the corresponding key component area), used to reflect the severity of the anomaly.

[0067] Specifically, the system distinguishes abnormal regions from normal regions through pixel-level classification or region segmentation, and uses connected component analysis to determine the complete range of abnormal regions. Subsequently, the system extracts the contours of abnormal regions to obtain the pixel coordinates of the abnormal region boundaries, and further calculates geometric feature parameters such as the area of ​​the abnormal region, contour length, size of the circumscribed rectangle, and region proportion.

[0068] In other embodiments, the separation of the abnormal region may also employ traditional segmentation methods based on threshold segmentation, edge detection, and region growing, or semantic segmentation and instance segmentation methods based on deep learning such as U-Net and Mask R-CNN.

[0069] S142. For the linear anomaly, the skeleton of the anomaly region is extracted, the skeleton coordinates are extracted, and the crack length, crack width and crack direction are calculated.

[0070] In this embodiment, skeleton coordinates refer to the set of coordinates of each pixel on the central skeleton with a single pixel width obtained after image thinning of a linear anomaly region with a certain width. The skeleton coordinates are used to achieve subsequent multi-cycle position matching. Crack length refers to the cumulative length of the pixels on the central skeleton along the skeleton extension direction, which is used to reflect the range of anomaly expansion. Crack width refers to the width of the anomaly region in the linear anomaly region, perpendicular to the central skeleton direction, which is used to describe the degree of structural damage. Crack direction refers to the angle between the main extension direction of the central skeleton and the preset reference direction, which is used to analyze the crack development trend.

[0071] Specifically, the system first uses edge enhancement and region segmentation methods to obtain linear anomaly regions (such as cracks, fissures, and broken edges). Then, it uses image thinning and skeleton extraction methods to convert the anomaly regions with a certain width into a central skeleton with a single pixel width. Finally, it calculates feature parameters such as crack length, crack width, crack direction, and crack distribution range based on the skeleton pixel coordinates.

[0072] In other embodiments, the skeleton extraction may also employ a median transformation or a distance transformation method.

[0073] S143. For the aforementioned status anomalies, calculate the door opening angle based on the positional relationship between the door area and the main body of the box, calculate the box tilt angle based on the edge contour of the box and the vertical reference direction, and determine the missing component based on the integrity detection of the key component area.

[0074] In this embodiment, the main body of the container refers to the fixed frame portion of the container facility excluding openable components such as the door; the door opening angle refers to the angle between the plane containing the edge of the door (or the line connecting the edges of the door) and the front of the main body of the container. In one embodiment, the state-related anomalies can be further divided into safety-related anomalies and posture-related anomalies: the safety-related anomalies correspond to the door area and the lock area, including the door open state and the lock state; the posture-related anomalies correspond to the box body and the base, including the box body tilt angle and the base state.

[0075] For lock malfunctions, the system performs integrity checks and status identification on the lock area, comparing the detection results of the current lock area with the normal lock structure template to determine whether the lock is missing, damaged, or abnormally opened. Missing or damaged locks are classified as component missing malfunctions. For base malfunctions, the system determines whether the base is loose, settled, or damaged based on the relative positional relationship between the base area and the ground reference line and the base outline. Loose or settled bases are classified as associated with abnormal box tilt, and damaged bases are classified as component missing malfunctions.

[0076] In one embodiment, the spatial offset of the door relative to the main body of the box can also be used to characterize the opening state of the door; the vertical reference direction refers to the direction of gravity or the direction of a reference line parallel to the direction of gravity, which can be determined in the image by calibration or gravity sensor data; the box tilt angle refers to the angle between the main extension direction of the box edge contour and the vertical reference direction, which is used to characterize the degree of deviation of the box posture from the normal vertical installation state.

[0077] Specifically, such as Figure 6 As shown, Figure 6 This is a schematic diagram of box tilt anomaly detection and tilt angle calculation provided in an embodiment of the present invention. Figure 6 In the image, label 101 indicates the bounding box of the box-shaped target detected by the system in the road environment video frame. The system extracts the box-shaped target region within this bounding box for tilt analysis (e.g., ...). Figure 6 (As shown in the enlarged view in the upper left corner); Label 102 indicates the box edge contour extracted by the system within the target area of ​​the box; Label 103 indicates the main extension direction line of the box edge obtained by fitting the box edge contour, which is obtained by fitting a straight line to the longitudinal edge of the box edge contour; Label 104 indicates the reference baseline, which in one embodiment is a reference line corresponding to the vertical reference direction, and in another embodiment can also be a horizontal reference line; Label 105 indicates the angle θ between the main extension direction line 103 of the box edge and the reference baseline 104. When the reference baseline 104 is a reference line corresponding to the vertical reference direction, the angle θ is the box tilt angle; when the reference baseline 104 is a horizontal reference line, the box tilt angle is the difference between 90° and the angle θ, that is, the angle by which the main extension direction of the box edge deviates from the vertical reference direction. When the box tilt angle is greater than a preset tilt angle threshold, it is determined that the box has an abnormal tilt, and the larger the value of the box tilt angle, the more serious the tilt of the box.

[0078] After completing the detection and feature extraction of various types of anomalies, the system associates the anomaly detection results with the absolute spatial location obtained in step S130 to establish an anomaly data record. This anomaly data record includes at least the container identifier, inspection time, absolute spatial location, anomaly type, anomaly region location (i.e., the location of the anomaly region in the container target image and its corresponding spatial location information), anomaly geometric features, and anomaly detection confidence. This anomaly data record is structured anomaly data containing anomaly type, anomaly region, anomaly degree, and spatial location. The system then transmits this structured anomaly data to step S150 for multi-period anomaly correlation analysis.

[0079] Through steps S141 to S143, multi-dimensional geometric features such as corrosion area, crack skeleton, and tilt angle are extracted for key components of the enclosure. This enables pixel-level quantitative description of features such as abnormal boundaries, crack skeleton, and tilt angle. Compared with the traditional method of only identifying the abnormality category, this method can more objectively and accurately reflect the abnormal state and damage degree of the enclosure, providing reliable data support for subsequent risk assessment.

[0080] S150. Based on the absolute spatial position, perform association matching on the abnormal geometric features of the same box target under different inspection cycles, assign box identifiers to the same box target, calculate abnormal evolution features, and generate abnormal evolution status codes.

[0081] In this embodiment, the inspection cycle refers to the time interval between two consecutive inspections of the same container facility, which can be a fixed cycle (such as once a week) or a cycle that is dynamically adjusted according to the risk status; the anomaly evolution characteristics refer to the characteristic parameters that characterize the changes and rates of change of the anomaly of the same container target during different inspection cycles; the anomaly evolution status code refers to the structured identifier obtained by classifying the current anomaly status of the container according to the anomaly development stage, which is used to uniformly describe the development stage of the anomaly.

[0082] Based on abnormal data acquired from multiple inspection cycles, the system performs correlation analysis on the abnormal change process of the same container target, establishes an abnormal time-series evolution model (i.e., a data model that describes the change process of abnormal geometric features in time sequence), extracts abnormal development trend features, and generates corresponding abnormal evolution status codes.

[0083] In one embodiment, step S150 described above may include steps S151 to S156.

[0084] S151. Based on the absolute spatial location and historical inspection data, determine whether the container targets in different inspection cycles belong to the same container target.

[0085] Specifically, the system acquires abnormal data recorded in different inspection cycles. This abnormal data includes at least the container's spatial location, inspection time, abnormality type, abnormal area location, abnormal geometric features, and abnormality detection confidence level. Since the image acquisition location, shooting angle, and environmental conditions for the same container may differ in different inspection cycles, the system matches container targets in different inspection cycles based on the container's spatial coordinates, target appearance features, structural features, and historical inspection records. When the spatial coordinate distance between container targets detected in two inspection cycles is less than a preset distance threshold, and the similarity between their appearance features and structural features is greater than a preset similarity threshold, they are determined to belong to the same container target. For example, the preset distance threshold can be set from 2 meters to 10 meters (e.g., 5 meters) depending on the positioning accuracy, and the preset similarity threshold can be set to 0.8.

[0086] S152. For the planar anomaly, match according to the boundary contour of the anomaly area; for the linear anomaly, match according to the skeleton coordinates and the crack direction; for the state anomaly, match according to the door opening angle and the box tilt angle; for the component missing anomaly, match according to the key component category, the missing part, and the component integrity detection result.

[0087] In this embodiment, key structural points refer to structural feature points on the enclosure that have a fixed relative positional relationship (such as enclosure corners, door hinge points, and lock installation points); spatial attitude parameters refer to parameters that characterize the spatial orientation and position of the enclosure or its components (such as enclosure tilt angle and door opening angle).

[0088] Specifically, for planar anomalies, the system matches anomaly regions across different inspection cycles based on their boundary contours, boundary pixel coordinates, and spatial relationships, treating anomalies in the same location as continuously evolving objects. For linear anomalies, the system matches crack anomalies across different inspection cycles based on skeleton coordinates, crack endpoint locations, and crack directions to determine whether cracks belong to the same continuously developing structural anomaly. For state-related anomalies, the system continuously correlates abnormal states such as door opening and enclosure tilting based on key structural point locations, spatial attitude parameters, and historical state information. Through these methods, a multi-cycle time-series data chain of enclosure anomalies is established, enabling the transformation from single anomaly detection results to long-term anomaly change processes.

[0089] S153. Project the abnormal regions extracted in each inspection cycle onto the geometric reference plane and normalize them.

[0090] Specifically, in order to eliminate the geometric distortion and quantization error caused by different shooting distances, shooting angles and imaging scaling ratios in different inspection cycles, the system performs spatial perspective correction and scale normalization on images from different inspection cycles: a local plane reference coordinate system for the front of the box is established using the corner points or edge contours of the box, and the abnormal areas extracted in each inspection cycle are projected onto a unified geometric reference plane through perspective transformation, thereby obtaining the normalized projected area of ​​the abnormal area. In one embodiment, the normalized projected area can also be converted into the corresponding real physical area by combining the shooting distance and imaging parameters.

[0091] S154. Calculate the abnormal evolution characteristics based on the changes in the abnormal geometric features during different inspection weeks. The abnormal evolution characteristics include the abnormal area growth rate, abnormal boundary expansion rate, crack length growth rate, crack width change rate, crack direction change, box attitude change rate, abnormal duration, and number of abnormal recurrences.

[0092] Specifically, regarding planar anomalies: Let the first... The normalized area of ​​the outlier region during the second inspection was: , No. The normalized area of ​​the outlier region during the second inspection was: The abnormal area growth rate is: in, This indicates the rate of change of the area of ​​the abnormal region per unit time. When the abnormal area continues to increase, it indicates that the abnormality is expanding, and the system can display the abnormal area growth rate. Compare with preset growth threshold and preset decrease threshold: When When the growth exceeds the preset threshold, it indicates that the abnormal region is expanding; when When the absolute value is less than the preset fluctuation threshold (i.e., the anomaly area remains basically unchanged), it indicates that the anomaly is in a stable state; when When the area is less than the preset reduction threshold (i.e., the anomaly area decreases), it indicates that the anomaly may have been repaired or the environmental impact may have lessened. When the anomaly area remains basically unchanged, it indicates that the anomaly is in a stable state. When the anomaly area decreases, it indicates that the anomaly may have been repaired or the environmental impact may have lessened.

[0093] For the matched boundary points and Calculate its spatial offset: ; The rate of anomaly boundary expansion is obtained based on the average change of multiple boundary points; at the same time, the main direction of anomaly expansion is determined based on the direction of the region with the greatest boundary change.

[0094] For linear anomalies: Let the crack lengths in the two inspection cycles be respectively... , The crack length growth rate is: ; Calculate the average crack width based on the distance to the crack region boundary: ; in, Let be the crack width at the i-th skeleton point; the crack width variation rate during different inspection cycles is: ; The crack direction changes as follows: ;in, , These represent the crack direction angles in the two inspection cycles.

[0095] For status-related anomalies: door opening angle The rate of change is: ; Box tilt angle The rate of change is: ; in, and Both belong to the rate of change of the container's posture, corresponding to the posture change situations of two types of abnormal states: abnormal opening of the container door and abnormal tilting of the container.

[0096] The duration of the anomaly is: Where n is the number of inspection cycles in which the anomaly persists. t is the time interval between adjacent inspection cycles; at the same time, the number of times the abnormality recurs is counted, that is, the cumulative number of times the same abnormality reappears after being dealt with.

[0097] In other embodiments, the calculation of the abnormal evolution characteristics may also employ sliding window regression to fit the abnormal change curve or a trend estimation method based on exponential smoothing.

[0098] S155. Classify the abnormal development stages according to the abnormal evolution characteristics, and generate the abnormal evolution status code. The abnormal evolution status code includes the first status code to the fifth status code in ascending order of the severity of the development stage.

[0099] Specifically, the system comprehensively considers indicators such as the abnormal area growth rate, crack propagation rate, box posture change rate, abnormal duration, and number of abnormal recurrences to classify abnormal states. For example, the classification rules are as follows: when no abnormality is detected, a first state code S0 is generated; when an abnormality exists and the growth rate of each abnormal evolution feature is lower than the corresponding preset lower threshold, a second state code S1 is generated, indicating a slight abnormality with a stable trend; when the growth rate of any abnormal evolution feature is between the preset lower threshold and the preset upper threshold and the abnormality persists, a third state code S2 is generated, indicating the abnormality persists and develops slowly; when the growth rate of any abnormal evolution feature exceeds the preset upper threshold, or the number of abnormal recurrences exceeds the preset number threshold, a fourth state code S3 is generated, indicating rapid expansion of the abnormality or continuous deterioration of the structural state; when a structural abnormality occurs in a critical component area (such as the box tilt angle exceeding the safety limit), a fifth state code S4 is generated, indicating the abnormality has reached a critical state requiring priority handling. The abnormal evolution state codes do not directly replace the risk assessment results but serve as structured identifiers describing the abnormal evolution stages, providing input information for subsequent health predictions.

[0100] In other embodiments, the abnormal evolution status code may also adopt a three-level, seven-level or other level of status division, and a status identifier in the form of letters, colors or other symbols.

[0101] S156. Construct an abnormal evolution feature vector by combining the abnormal evolution features and the abnormal evolution status code.

[0102] In this embodiment, the abnormal evolution feature vector refers to a multi-dimensional feature vector formed by arranging various abnormal evolution features and abnormal evolution status codes in a preset order. For example, the abnormal evolution feature vector is represented as follows: ;in, Let D be the rate of increase of the abnormal area and D be the rate of expansion of the abnormal boundary. This represents the crack length growth rate. The crack width variation rate, Due to the change in crack direction, The rate of change of the door opening angle. The rate of change of the box tilt angle ( and Together they constitute the box's attitude change rate, T is the duration of the anomaly, N is the number of times the anomaly recurred, and C is the anomaly evolution status code. When the box does not have a certain type of anomaly, the corresponding vector component is set to zero or a preset default value.

[0103] Through steps S151 to S156, a correlation mechanism for multi-cycle inspection data was established, enabling continuous tracking and evolution analysis of abnormal states. This allows for the differentiation between short-term, occasional abnormalities and continuously deteriorating abnormalities, overcoming the limitations of existing technologies that only judge based on single inspection results, and providing a time-series data foundation for health trend prediction.

[0104] S160. Based on the abnormal evolution characteristics, the abnormal evolution status code, and historical inspection data, predict the health status change trend of the container target, make corrections in conjunction with environmental influencing factors, calculate the container health index and comprehensive risk value, and determine the risk level.

[0105] In this embodiment, historical inspection data refers to the inspection records of each inspection cycle prior to the current inspection cycle stored in the container's health file, including the results of each abnormal detection, abnormal evolution characteristics, maintenance records, and corresponding environmental information; the health status change trend refers to the expected evolution direction of the container's health status in future inspection cycles, including a continuous degradation trend, a stable operating state, and a recovery state; the container health index refers to a numerical indicator obtained by uniformly and quantitatively evaluating the overall health level of the container; the comprehensive risk value refers to a quantitative indicator that characterizes the magnitude of the container's operational risk after integrating the container's current abnormal state, development trend, and environmental impact; and the risk level refers to the classification result of the container's risk status based on the threshold range of the comprehensive risk value.

[0106] like Figure 7 As shown, in one embodiment, step S160 may include steps S161-S162 as well as environmental correction, comprehensive risk assessment and prediction credibility assessment.

[0107] In one embodiment, step S160 described above may include S161 to S162.

[0108] S161. Using the abnormal evolution feature vector as input and combining it with the abnormal change patterns in historical inspection data, predict the health status change trend of the container target in future inspection cycles to obtain the prediction result.

[0109] In this embodiment, specifically, the system constructs a current health status feature set (i.e., a feature set describing the current abnormality level and historical trend of the container) based on the abnormal evolution feature vector and the abnormal evolution status code, combined with the container's historical inspection records, and performs health status prediction based on the container health status feature set.

[0110] In this embodiment, the prediction result refers to the judgment information about the future health status of the container output by the health status prediction model, including the trend category and the predicted values ​​of various abnormal evolution characteristics. The health status prediction model includes a container health decline model, which is a mathematical model describing the gradual degradation of the container's health status as an abnormal evolution process. It can employ time series analysis methods (such as autoregressive integral moving average models), statistical prediction methods (such as grey prediction models), machine learning methods (such as recurrent neural networks, gradient boosting decision trees), or other methods capable of predicting state trends. The health status prediction model is trained or calibrated using historical inspection data accumulated in the container's health records. By learning from historical abnormal evolution patterns, it determines whether the current abnormality has a continuous development trend. Further, when the prediction result indicates a continuous degradation trend, the system extrapolates the remaining time required for the abnormal evolution characteristics to reach a preset risk threshold based on the growth rate of the current abnormal evolution characteristics, which is taken as the remaining safe lifespan of the container target. When the remaining safe lifespan is less than the preset lifespan threshold, the system increases the inspection frequency of the corresponding container and generates a disposal prompt in advance, realizing abnormal development trend analysis and remaining safe lifespan prediction.

[0111] In this embodiment, historical maintenance information refers to the time, content, and post-maintenance status information of each maintenance performed on the enclosure, as recorded in the enclosure's health record.

[0112] Specifically, when the prediction result is that the degree of abnormality continues to increase, the rate of abnormality expansion increases, or the development stage represented by the abnormality evolution status code continues to deteriorate, it is determined that it is in a continuous degradation trend; when the change in the degree of abnormality is less than the preset change threshold and the abnormality evolution status code remains stable, it is determined that it is in a stable operating state; when the degree of abnormality decreases or the abnormal area shrinks, it is determined to enter the recovery state by combining the historical maintenance information of the box target. That is, only when the historical maintenance information indicates that the abnormal area has been repaired will the decrease in the degree of abnormality be determined as the recovery state, so as to eliminate misjudgment caused by detection error.

[0113] Through the above prediction process, the transformation from "current anomaly detection" to "future health trend prediction" is realized, which helps to identify potential risks in advance and improves the initiative of container operation and maintenance management.

[0114] S162. The abnormal evolution features and the prediction results in the abnormal evolution feature vector are weighted and fused to obtain the box health index.

[0115] The environmental impact factors include road grade, traffic flow, rainfall, air humidity, coastal salt spray environment, industrial pollution, and obstruction. When the container target is in a high humidity environment or salt spray environment, the influence weight of the planar anomalies corresponding to rust and corrosion is increased; when the traffic flow in the area where the container target is located is greater than a preset traffic flow threshold, the influence weight of the linear anomalies and the state-type anomalies is increased. The container health index is updated based on the corrected impact weights of each anomaly, and the overall risk value is updated again using the updated container health index to obtain the environmental correction result.

[0116] In this embodiment, if the health index of the enclosure is lower than the preset safety threshold, the corresponding risk level is increased or a corresponding warning message is generated based on the health index of the enclosure, the comprehensive risk value and the preset risk threshold.

[0117] In one embodiment, calculating the health index and comprehensive risk value of the enclosure, and determining the risk level, includes: The comprehensive risk value is calculated based on the health index of the enclosure, the prediction result, and the environmental correction result, and the risk level is divided into normal, low risk, medium risk, high risk, and severe risk according to the preset risk threshold. The prediction confidence level is calculated based on historical prediction error, anomaly detection confidence level, and historical data completeness. The prediction confidence level is not included in the calculation of the comprehensive risk value. When the prediction confidence level is lower than a preset confidence level threshold, the corresponding prediction result is marked to indicate the need for review.

[0118] Specifically, the system employs a multi-indicator fusion approach. First, each abnormal evolution characteristic is normalized into a health deduction item, then weighted calculations are performed. Planar abnormal features are used to evaluate the aging and corrosion degree of the enclosure surface; linear abnormal features are used to evaluate the degree of structural damage development; state-related abnormal features are used to evaluate the enclosure's installation stability and structural condition; component missingness is used to evaluate the enclosure's safety status; the duration and frequency of abnormal occurrences are used to evaluate the long-term impact of the abnormality; and prediction results are used to assist in judging future health trends. When the abnormal area continues to expand, the crack growth rate increases, the posture changes significantly, or the duration of the abnormality increases, the enclosure health index decreases; when the abnormality is repaired and restored, the detection deviation is confirmed by re-inspection, or a new rule for the decay or maintenance of a stable state is defined, the enclosure health index increases.

[0119] In other embodiments, the health index of the enclosure can also be calculated using the analytic hierarchy process (AHP), entropy weight method, fuzzy comprehensive evaluation method, or a health assessment model based on a neural network. The enclosure health index provides a unified quantitative description of the enclosure's health status throughout its entire lifecycle.

[0120] Environmental Correction: Considering the different impacts of various operating environments on the development of anomalies in the enclosure, the system introduces an environmental impact correction mechanism. The environmental impact factors include road grade, traffic flow, rainfall, air humidity, salt spray environment, industrial pollution, and obstruction. These factors can be obtained through meteorological data interfaces, road management databases, or on-site sensors. Specifically, when the enclosure target is in a high-humidity or salt spray environment, the influence weight of the areal anomalies corresponding to rust and corrosion is increased; when the traffic flow in the area where the enclosure target is located exceeds a preset flow threshold, the influence weight of the linear anomalies and state-related anomalies is increased; for areas with stable environmental conditions, the correction magnitude of the environmental impact factors is reduced; for example, the increase in influence weight can be 1.2 to 1.5 times the original weight. The enclosure health index is updated according to the corrected influence weights of each anomaly, and the comprehensive risk value is updated again using the updated enclosure health index to obtain the environmental correction result. Through environmental correction, the matching degree between the health status evaluation results and the actual operating environment is improved.

[0121] Comprehensive risk assessment: The system calculates the comprehensive risk value based on the container health index, the prediction results, and the environmental correction results; In one embodiment, when calculating the comprehensive risk value, the abnormal evolution feature vector, the abnormal evolution status code, and the facility importance can be further integrated; wherein, the facility importance refers to a level parameter that is pre-set according to the functional category, number of service objects, and scope of fault impact of the enclosure facility. For example, the facility importance of an enclosure facility that supplies power to traffic signal control is higher than that of a general lighting control enclosure.

[0122] Based on preset risk thresholds, the risk levels are divided into five categories: normal, low risk, medium risk, high risk, and severe risk. Wherein: normal indicates that the container is operating stably with no obvious abnormalities; low risk indicates minor abnormalities requiring continued monitoring; medium risk indicates that the abnormality shows a tendency to worsen, and on-site inspection is recommended; high risk indicates that the abnormality has affected the normal operation of the container and requires priority handling; severe risk indicates a significant safety hazard requiring immediate action. In other embodiments, the risk level classification can also employ adaptive partitioning using cluster analysis or a fuzzy membership-based hierarchical method.

[0123] Prediction credibility assessment: The system calculates prediction credibility based on historical prediction errors, anomaly detection confidence, and the completeness of historical data; In one embodiment, the prediction reliability can be further calculated by combining the continuity of inspection data and the degree of influence of environmental correction.

[0124] It should be noted that the prediction credibility is only used as an auxiliary indicator to evaluate the reliability of the prediction results, and it does not participate in the calculation of the comprehensive risk value or affect the risk level classification. When the prediction credibility is higher than or equal to the preset credibility threshold, it indicates that the prediction result has high reliability and can be directly used for subsequent early warning and operation and maintenance decisions; when the prediction credibility is lower than the preset credibility threshold, the corresponding prediction result is marked to indicate review, and the system prompts operation and maintenance personnel to conduct manual review or increase the frequency of inspections.

[0125] Finally, the system outputs the health status prediction results of the enclosure, which include at least the health status of the enclosure, the health index of the enclosure, the trend of health changes, the comprehensive risk value, the risk level, the abnormal evolution status code, and the prediction confidence. The above results are then passed to step S170.

[0126] Step S160 integrates multi-dimensional information such as abnormal evolution trends, environmental impact factors, and facility importance to conduct a comprehensive risk assessment, overcoming the problem of inaccurate assessment caused by alarms based on single abnormal features in existing technologies. At the same time, it decouples prediction reliability from risk value, ensuring the objectivity of risk level classification and improving the accuracy of risk assessment and the pertinence of early warning.

[0127] S170. Generate early warning information and operation and maintenance task information according to the risk level, and update the container health record of the container target.

[0128] In this embodiment, early warning information refers to information generated based on risk levels to prompt maintenance personnel to take corresponding measures, including early warning level, container identification, anomaly description, and handling suggestions; maintenance task information refers to task description information used to guide maintenance personnel in carrying out on-site handling; container health records refer to dynamic data archives established for each container facility, recording its health information throughout its entire life cycle, including target spatial location, characteristics of previous anomalies, risk level, early warning information, and maintenance results. By generating differentiated early warning information and maintenance tasks according to risk levels, the classification and handling of abnormal events and maintenance scheduling are realized, improving the scheduling efficiency of maintenance resources and the timeliness of anomaly handling.

[0129] like Figure 8As shown, specifically, the system generates corresponding early warning information for enclosures with different risk levels based on preset early warning rules: when the risk level is normal, the current health status is recorded and periodic inspections continue; when the risk level is low risk, a monitoring reminder is generated, prompting maintenance personnel to conduct continuous observation; when the risk level is medium risk, an inspection review task is generated, reminding maintenance personnel to further confirm the abnormal situation; when the risk level is high risk or severe risk, a key handling early warning is generated, prompting timely maintenance or safety handling. Simultaneously, the system combines the abnormal evolution status code to generate differentiated early warning strategies: when enclosures have the same risk level but different abnormal evolution status codes, a higher task priority is generated for enclosures with more severe development stages represented by the abnormal evolution status code; when the abnormal evolution status code of the same enclosure continuously upgrades during consecutive inspection cycles, the system automatically increases the early warning level of that enclosure and shortens its inspection cycle, realizing the classification and handling of abnormal events, intelligent early warning, and maintenance scheduling, improving the accuracy of risk assessment and the pertinence of early warnings.

[0130] The operation and maintenance task information is generated, which includes the container identifier, the absolute spatial location, the anomaly type, the risk level, the suggested handling measures, and the task priority. The absolute spatial location is derived from the container positioning result in step S130, enabling operation and maintenance personnel to quickly locate abnormal facilities and improve maintenance efficiency. The task priority can be determined according to the risk level and the importance of the facility. The higher the risk level and the higher the importance of the facility, the higher the operation and maintenance task priority of the container.

[0131] After completing on-site inspections or repairs, the system acquires maintenance feedback information, including repair completion status and anomaly resolution status. In one embodiment, this feedback also includes information on replaced components, repair time, and on-site verification results. The system associates the maintenance feedback information with the original anomaly records: when an anomaly is resolved after repair, the impact of the corresponding anomaly is reduced and the enclosure health index is updated; when an anomaly persists or continues to develop, historical anomaly records are retained and the level of attention for subsequent inspections is increased. Through maintenance feedback, a data loop is achieved between anomaly detection results and actual maintenance results.

[0132] Furthermore, the system writes new inspection data, abnormal change processes, and the operation and maintenance feedback information into the container health record to update the abnormal evolution record of the container target; the new data can also be used to optimize health status evaluation parameters, correct the prediction results of the health status prediction model, and improve the accuracy of subsequent risk assessment. By continuously accumulating container operation data, the system can achieve dynamic improvement in the container health status management capability.

[0133] Through step S170, the target spatial location, anomaly characteristics, risk level, early warning information and operation and maintenance results are continuously correlated, forming a closed-loop management of the entire life cycle from anomaly detection, risk warning, maintenance and disposal to status update. This provides complete data support for facility health status traceability, operation and maintenance decision-making and long-term management, and improves the level of intelligent management of urban infrastructure.

[0134] The aforementioned method for detecting abnormal conditions of enclosures based on dynamic inspection achieves precise enclosure positioning by fusing vehicle-mounted video and positioning information to perform spatial line-of-sight intersection. It extracts refined geometric features of key component anomalies, calculates anomaly evolution features based on absolute spatial location and multi-period data, and generates status codes. It combines historical data and environmental factors to predict health trends and assess risk levels, thereby generating intelligent early warnings and maintenance tasks and dynamically updating health records. This solves problems such as inaccurate positioning, insufficient feature quantification, lack of trend tracking, and fragmented data management in dynamic inspections, significantly improving the accuracy of anomaly detection and inspection efficiency, and providing full lifecycle data support for intelligent facility operation and maintenance.

[0135] Figure 8 This is a schematic block diagram of a container abnormality detection system 300 based on dynamic inspection, provided in an embodiment of the present invention. Figure 8 As shown, corresponding to the above-described method for detecting abnormal states of enclosures based on dynamic inspection, the present invention also provides a system 300 for detecting abnormal states of enclosures based on dynamic inspection. This system 300 includes a unit for executing the above-described method for detecting abnormal states of enclosures based on dynamic inspection, and the system can be configured in a server. Specifically, please refer to... Figure 8 The container abnormality detection system 300 based on dynamic inspection includes a data acquisition unit 301, a target tracking unit 302, a spatial positioning unit 303, an anomaly detection unit 304, an evolution analysis unit 305, a risk assessment unit 306, and an early warning closed-loop unit 307.

[0136] The data acquisition unit 301 is used to acquire road environment videos collected by the patrol vehicle during its operation, as well as vehicle pose information corresponding to each video frame; the target tracking unit 302 is used to perform box target detection and cross-frame tracking on the road environment videos to obtain a continuous observation sequence; the spatial positioning unit 303 is used to perform spatial line intersection on the observation lines of the same box target in different video frames in the continuous observation sequence, based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment videos, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, after converting them to the same reference coordinate system, and converting the relative spatial position obtained by the intersection into an absolute spatial position in combination with the vehicle pose information; the anomaly detection unit 304 is used to detect the target target in the road environment videos. Anomaly detection is performed on key component areas of the container target, and abnormal geometric features are extracted; the evolution analysis unit 305 is used to perform correlation matching of the abnormal geometric features of the same container target under different inspection cycles based on the absolute spatial location, assign container identifiers to the same container target, calculate abnormal evolution features, and generate abnormal evolution status codes; the risk assessment unit 306 is used to predict the health status change trend of the container target based on the abnormal evolution features, the abnormal evolution status codes, and historical inspection data, make corrections in combination with environmental impact factors, calculate the container health index and comprehensive risk value, and determine the risk level; the early warning closed-loop unit 307 is used to generate early warning information and operation and maintenance task information based on the risk level, and update the container health record of the container target.

[0137] In one embodiment, the spatial positioning unit 303 includes: The observation line calculation subunit is used to establish a camera projection model based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, and calculate the observation lines corresponding to the feature points of the box target in different video frames; the synchronization subunit is used to convert the observation lines in different video frames to the same reference coordinate system based on the vehicle pose information corresponding to each video frame and the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle; the intersection calculation subunit is used to perform intersection calculations on the observation lines of the same feature point after conversion to the same reference coordinate system to obtain multiple relative spatial positions; the fusion optimization subunit is used to fuse and optimize the multiple relative spatial positions based on the target detection confidence and spatial calculation error of each video frame; the position transformation subunit is used to transform the fused and optimized relative spatial positions to the geographic coordinate system based on the installation position parameters, the installation attitude parameters and the vehicle pose information to obtain the absolute spatial position.

[0138] In one embodiment, the anomaly detection unit 304 is used to, for the planar anomaly, separate the abnormal region from the normal region, extract the boundary contour of the abnormal region, and calculate the area and proportion of the abnormal region; for the linear anomaly, extract the skeleton of the abnormal region, extract the skeleton coordinates, and calculate the crack length, crack width, and crack direction; for the state-type anomaly, calculate the door opening angle based on the positional relationship between the door area and the main body of the box, calculate the box tilt angle based on the edge contour of the box and the vertical reference direction, and determine the missing component based on the integrity detection of the key component area.

[0139] In one embodiment, the evolution analysis unit 305 includes: The judgment subunit is used to determine whether the box targets in different inspection cycles belong to the same box target based on the absolute spatial position and historical inspection data; the matching subunit is used to match the planar anomaly based on the boundary contour of the anomaly area; for the linear anomaly, it is used to match based on the skeleton coordinates and the crack direction; for the state anomaly, it is used to match based on the door opening angle and the box tilt angle; for the component missing anomaly, it is used to match based on the key component category, the missing part, and the component integrity detection result; the normalization subunit is used to project and map the anomaly area extracted in each inspection cycle onto the geometric reference plane for normalization. The feature calculation subunit is used to calculate the anomaly evolution features based on the changes in the anomaly's geometric features during different inspection cycles. The anomaly evolution features include the anomaly area growth rate, the anomaly boundary expansion rate, the crack length growth rate, the crack width change rate, the crack direction change, the box attitude change rate, the anomaly duration, and the number of anomaly recurrences. The status code generation subunit is used to classify the anomaly development stages based on the anomaly evolution features and generate the anomaly evolution status codes. The anomaly evolution status codes include the first to the fifth status codes in ascending order of the severity of the development stages. The vector construction subunit is used to construct an anomaly evolution feature vector by combining the anomaly evolution features and the anomaly evolution status codes.

[0140] In one embodiment, the risk assessment unit 306 includes: The prediction subunit is used to predict the health status trend of the container target in future inspection cycles by taking the abnormal evolution feature vector as input and combining the abnormal change patterns in historical inspection data, so as to obtain the prediction result. The weighted fusion subunit is used to perform weighted fusion of each abnormal evolution feature in the abnormal evolution feature vector and the prediction result to obtain the container health index. The environmental influencing factors include road grade, traffic flow, rainfall, air humidity, salt spray environment, industrial pollution, and shading. The correction subunit is used to increase the influence weight of the areal anomalies corresponding to rust and corrosion when the container target is in a high humidity environment or salt spray environment; and to increase the influence weight of the linear anomalies and the state-type anomalies when the traffic flow in the area where the container target is located is greater than a preset flow threshold. The container health index is updated according to the corrected influence weights of each anomaly, and the comprehensive risk value is updated again using the updated container health index to obtain the environmental correction result. The system includes a division subunit, used to calculate the comprehensive risk value based on the container health index, the prediction result, and the environmental correction result, and to classify the risk level into normal, low risk, medium risk, high risk, and severe risk according to a preset risk threshold; and a credibility calculation subunit, used to calculate the prediction credibility based on historical prediction error, anomaly detection confidence, and historical data completeness, wherein the prediction credibility is not included in the calculation of the comprehensive risk value; and when the prediction credibility is lower than the preset credibility threshold, the corresponding prediction result is marked to indicate a review.

[0141] In one embodiment, the early warning closed-loop unit 307 is used to record the current health status when the risk level is normal; generate a attention reminder or inspection and review task when the risk level is low or medium risk; generate a key handling warning when the risk level is high or severe risk; generate maintenance task information, which includes the enclosure identifier, the absolute spatial location, the anomaly type, the risk level, suggested handling measures, and the task priority; obtain maintenance feedback information including maintenance completion status and anomaly elimination status; update the enclosure health index when the anomaly is eliminated after maintenance; retain historical anomaly records and increase the subsequent inspection attention level when the anomaly still exists; and write new inspection data, anomaly change process, and maintenance feedback information into the enclosure health file.

[0142] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned container abnormality detection system 300 based on dynamic inspection and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0143] The aforementioned container abnormality detection system 300 based on dynamic inspection can be implemented as a computer program, which can, for example... Figure 9 It runs on the computer device shown.

[0144] Please see Figure 9 , Figure 9 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0145] See Figure 9 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0146] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a dynamic inspection-based enclosure abnormality detection method.

[0147] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0148] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a method for detecting abnormal states of the enclosure based on dynamic inspection.

[0149] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] The processor 502 is used to run the computer program 5032 stored in the memory to implement all the steps of the container abnormality detection method based on dynamic inspection.

[0151] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0152] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0153] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform all the steps of the dynamic inspection-based enclosure anomaly detection method.

[0154] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0155] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0156] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each unit is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0157] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the system of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0158] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0159] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting abnormal states of enclosures based on dynamic inspection, characterized in that, include: Acquire road environment videos collected by the patrol vehicle during its operation, as well as vehicle pose information corresponding to each video frame; Box target detection and cross-frame tracking are performed on the road environment video to obtain a continuous observation sequence; Based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, the observation lines of the same box target in different video frames in the continuous observation sequence are converted to the same reference coordinate system and then spatial line intersection is performed. The relative spatial position obtained by the intersection is converted into an absolute spatial position by combining the vehicle pose information. Anomaly detection is performed on the key component areas of the box target, and abnormal geometric features are extracted; Based on the absolute spatial location, the abnormal geometric features of the same container target under different inspection cycles are associated and matched, and a container identifier is assigned to the same container target. The abnormal evolution features are calculated and an abnormal evolution status code is generated. Based on the abnormal evolution characteristics, the abnormal evolution status codes, and historical inspection data, the health status change trend of the container target is predicted, and corrections are made in conjunction with environmental influencing factors. The container health index and comprehensive risk value are calculated, and the risk level is determined. Based on the risk level, early warning information and operation and maintenance task information are generated, and the health record of the container target is updated.

2. The method for detecting abnormal states of a container based on dynamic inspection according to claim 1, characterized in that, Based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, spatial line-of-sight intersection is performed on the same box target in different video frames in the continuous observation sequence after converting the observation lines to the same reference coordinate system. The relative spatial position obtained from the intersection is then converted into an absolute spatial position by combining the vehicle pose information. This includes: A camera projection model is established based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, and the observation line of view corresponding to the feature points of the box target in different video frames is calculated. Based on the vehicle pose information corresponding to each video frame and the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, the observation line of sight under different video frames is converted to the same reference coordinate system; Intersection calculations are performed on the observation lines of the same feature point after transformation to the same reference coordinate system to obtain multiple relative spatial positions; The relative spatial positions of each video frame are fused and optimized based on the target detection confidence and spatial calculation error. Based on the installation location parameters, the installation posture parameters, and the vehicle pose information, the fused and optimized relative spatial position is transformed to a geographic coordinate system to obtain the absolute spatial position.

3. The method for detecting abnormal states of a container based on dynamic inspection according to claim 1, characterized in that, The anomalies include planar anomalies, linear anomalies, and condition anomalies. Planar anomalies include rust, paint peeling, corrosion, and dirt. Linear anomalies include cracks and fissures. Condition anomalies include open cabinet doors, tilted cabinets, and missing parts. The extraction of abnormal geometric features includes: For the somatic anomalies, the abnormal region is separated from the normal region, the boundary contour of the abnormal region is extracted, and the area and proportion of the abnormal region are calculated. For the linear anomaly, the skeleton of the anomaly region is extracted, the skeleton coordinates are extracted, and the crack length, crack width and crack direction are calculated. For the aforementioned status anomalies, the door opening angle is calculated based on the positional relationship between the door area and the main body of the box, the box tilt angle is calculated based on the edge contour of the box and the vertical reference direction, and the missing component is determined based on the integrity detection of the key component area.

4. The method for detecting abnormal states of a container based on dynamic inspection according to claim 3, characterized in that, The association and matching of the abnormal geometric features of the same container target under different inspection cycles based on the absolute spatial location includes: Based on the absolute spatial location and historical inspection data, determine whether the container targets in different inspection cycles belong to the same container target; For the planar anomaly, matching is performed based on the boundary contour of the anomaly region; for the linear anomaly, matching is performed based on the skeleton coordinates and the crack direction; for the state-related anomaly, matching is performed based on the door opening angle and the box body tilt angle; for the component missing anomaly, matching is performed based on the key component category, the missing part, and the component integrity detection results. The abnormal regions extracted in each inspection cycle are projected onto a geometric reference plane and normalized.

5. The method for detecting abnormal states of a container based on dynamic inspection according to claim 4, characterized in that, The calculation of anomalous evolution characteristics and the generation of anomalous evolution status codes include: The abnormal evolution characteristics are calculated based on the changes in the abnormal geometric features during different inspection weeks. The abnormal evolution characteristics include the abnormal area growth rate, the abnormal boundary propagation rate, the crack length growth rate, the crack width change rate, the crack direction change, the box attitude change rate, the abnormal duration, and the number of abnormal recurrences. The abnormal development stages are classified according to the abnormal evolution characteristics, and the abnormal evolution status codes are generated. The abnormal evolution status codes include the first status code to the fifth status code in ascending order of the severity of the development stages. The abnormal evolution features and the abnormal evolution status codes are used to construct an abnormal evolution feature vector.

6. The method for detecting abnormal states of a container based on dynamic inspection according to claim 5, characterized in that, The step of predicting the health status change trend of the container target based on the abnormal evolution characteristics, the abnormal evolution status code, and historical inspection data includes: Using the abnormal evolution feature vector as input and combining it with the abnormal change patterns in historical inspection data, the health status change trend of the container target in future inspection cycles is predicted to obtain the prediction result. The abnormal evolution features and the prediction results in the abnormal evolution feature vector are weighted and fused to obtain the box health index.

7. The method for detecting abnormal states of a container based on dynamic inspection according to claim 6, characterized in that, The correction based on environmental impact factors includes: The environmental impact factors include road grade, traffic flow, rainfall, air humidity, salt spray environment, industrial pollution, and obstruction. When the container target is in a high humidity environment or salt spray environment, the influence weight of the planar anomalies corresponding to rust and corrosion is increased; when the traffic flow in the area where the container target is located is greater than a preset traffic flow threshold, the influence weight of the linear anomalies and the state-type anomalies is increased. The container health index is updated based on the corrected impact weights of each anomaly, and the overall risk value is updated again using the updated container health index to obtain the environmental correction result.

8. The method for detecting abnormal states of a container based on dynamic inspection according to claim 7, characterized in that, The calculation of the enclosure's health index and comprehensive risk value, and the determination of the risk level, include: The comprehensive risk value is calculated based on the health index of the enclosure, the prediction result, and the environmental correction result, and the risk level is divided into normal, low risk, medium risk, high risk, and severe risk according to the preset risk threshold. The prediction confidence level is calculated based on historical prediction error, anomaly detection confidence level, and historical data completeness. The prediction confidence level is not included in the calculation of the comprehensive risk value. When the prediction confidence level is lower than a preset confidence level threshold, the corresponding prediction result is marked to indicate the need for review.

9. The method for detecting abnormal states of a container based on dynamic inspection according to claim 8, characterized in that, The process of generating early warning information and maintenance task information based on the risk level, and updating the container health record of the container target, includes: When the risk level is normal, record the current health status; when the risk level is low or medium risk, generate a reminder or inspection and review task; when the risk level is high or severe risk, generate a key handling warning. Generate operation and maintenance task information, which includes the container identifier, the absolute spatial location, the anomaly type, the risk level, the suggested handling measures, and the task priority; Obtain maintenance feedback information including maintenance completion status and anomaly elimination status; update the health index of the enclosure when the anomaly is eliminated after maintenance; retain historical anomaly records and increase the attention level of subsequent inspections when the anomaly still exists. New inspection data, abnormal changes, and maintenance feedback information are written into the health record of the enclosure.

10. A container abnormality detection system based on dynamic inspection, characterized in that, include: The data acquisition unit is used to acquire road environment videos collected by the patrol vehicle during its operation, as well as vehicle pose information corresponding to each video frame. The target tracking unit is used to perform box target detection and cross-frame tracking on the road environment video to obtain a continuous observation sequence; The spatial positioning unit is used to perform spatial line intersection on the observation lines of the same box target in different video frames in the continuous observation sequence, based on the internal imaging parameters of the vehicle-mounted camera that collects the road environment video, the installation position parameters and installation attitude parameters of the vehicle-mounted camera relative to the patrol vehicle, and the relative spatial position obtained by the intersection is converted to the same reference coordinate system after the observation lines of the same box target in different video frames are converted to the absolute spatial position by combining the vehicle pose information. An anomaly detection unit is used to detect anomalies in key component areas of the box target and extract abnormal geometric features; The evolution analysis unit is used to perform correlation matching of the abnormal geometric features of the same box target under different inspection cycles based on the absolute spatial position, assign box identifiers to the same box target, calculate abnormal evolution features, and generate abnormal evolution status codes. The risk assessment unit is used to predict the health status change trend of the container target based on the abnormal evolution characteristics, the abnormal evolution status code and historical inspection data, make corrections in combination with environmental impact factors, calculate the container health index and comprehensive risk value, and determine the risk level. The early warning closed-loop unit is used to generate early warning information and operation and maintenance task information according to the risk level, and update the container health record of the container target.

Images