Intelligent park alarm method and system based on intelligent gateway

Abstract

The invention discloses a smart park alarm method and system based on an intelligent gateway, and the method comprises the following steps: accessing a heterogeneous terminal through a multi-protocol southbound interface, generating a standardized event set, and building a spatial topology table; performing quality evaluation on the standardized event set to form a sequential cleaning event stream; target detection is carried out based on the sequential cleaning event flow, and a semantic event flow is output; performing space-time matching on the semantic event flow to obtain a fusion event set; according to a preset grading standard, a grading warning set is generated, and a linkage instruction set is generated; scheduling and executing the execution end according to the linkage instruction set to form a processing state feedback; integrating the grading warning set and the processing state feedback into an evidence packet, and reporting the evidence packet in parallel through the main and standby reporting links; and archiving and indexing the evidence packet and the reported result, generating an auditable record, and finishing synchronous updating with an upper platform. According to the invention, the smart park alarm method and system are realized.

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IoT) technology, and in particular to a smart park alarm method and system based on a smart gateway. Background Technology

[0002] Existing technologies in the field of smart park alarms largely rely on direct connection of front-end devices to the platform or forwarding via lightweight edge gateways, facing problems such as diverse protocols and inconsistent data interpretation. Different vendors' devices employ various southbound protocols, and the platform side often uses adapters for compatibility, leading to differences in field semantics, time stamps, and location descriptions for the same event, resulting in data silos and duplicate alarms. Video and non-video data have long been processed separately, lacking unified cleaning and quality assessment at the edge; time synchronization, jitter reduction, and noise reduction often remain at simple threshold filtering, making it difficult to guarantee the stability and consistency of subsequent identification and linkage.

[0003] At the event understanding level, traditional approaches often rely on single-point thresholds or device-level rules for triggering, lacking spatiotemporal constraints and pattern aggregation based on the spatial topology of the park. This makes it difficult to identify combined anomalies across regions and channels, often resulting in false alarms and missed alarms. Regarding coordinated response, existing systems generally centralize action distribution to the platform for scheduling, lacking orchestration and acknowledgment loops for dependencies and mutual exclusions. This leads to issues such as concurrent conflicts between multiple devices, duplicate executions, or prolonged periods without acknowledgments and retrying, making it difficult to quantify, archive, and audit the status of responses.

[0004] Regarding reporting and reliable transmission, existing solutions mostly employ fixed redundancy or symmetric erasure coding. Once the uplink experiences jitter or fragment loss, the system cannot adaptively adjust the degree distribution based on link health, nor does it perform coverage orthogonality and rank control for primary and backup links, resulting in wasted redundancy and low recovery efficiency. In weak network environments, common issues include coarse-grained breakpoint resumption and idempotent control, incomplete cross-window confirmation, and incomplete uploading of evidence data and logs, affecting event closure.

[0005] Therefore, how to provide a smart park alarm method and system based on a smart gateway is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a smart park alarm method and system based on a smart gateway. This invention achieves multi-protocol access and semantic alignment through a smart gateway, completes time-series cleaning and target detection, performs spatiotemporal fusion based on spatial topology and classifies according to preset standards, drives linkage and generates handling receipts, and reports evidence packages in parallel via primary and backup links. Archiving and auditing are simultaneously uploaded to the upper-level platform, significantly improving the real-time performance, reliability and traceability of park alarms.

[0007] A smart campus alarm method based on a smart gateway according to an embodiment of the present invention includes the following steps:

[0008] The multi-protocol southbound interface of the smart gateway connects to heterogeneous terminals, completes data parsing and semantic alignment, generates standardized event sets, and establishes a spatial topology table.

[0009] The standardized event set is time-synchronized, de-jittered, denoised, and quality-evaluated to form a time-series cleaned event stream and written into the local time-series database.

[0010] Target detection, tracking, and behavior recognition are performed based on a time-series cleansing event stream, and a semantic event stream is output.

[0011] Using the spatial topology table as a constraint, spatiotemporal matching, pattern matching, and event aggregation are performed on the semantic event stream to obtain a fused event set.

[0012] Based on the preset grading standards, the fusion event set is graded and evaluated, a graded warning set is generated, and a linkage instruction set is produced.

[0013] According to the linkage instruction set, the execution end is scheduled and executed under the constraints of dependency and mutual exclusion, and the status feedback is generated.

[0014] The graded warning set and the handling status feedback are integrated into an evidence package and reported in parallel through the primary and backup reporting links. The reporting process adopts online fountain code asymmetric redundancy, combined with cross-link orthogonal pairing coding to suppress rank degradation, and uses sequence number and idempotent identifier to implement confirmation, retransmission and breakpoint resume control.

[0015] Archive and index the evidence package and reported results, generate auditable records, and complete the synchronization update with the upper-level platform.

[0016] Optionally, the generation of the standardized event set and the establishment of the spatial topology table specifically include:

[0017] Configure the name, version, port and medium of the southbound protocol list on the pre-deployed smart gateway, scan and discover heterogeneous terminals within the park, complete identity authentication and address allocation, establish sessions according to terminals and protocols and register session registry, open raw data receiving queue and buffer, set packet loss alarm and rate limiting policies, and output raw data stream.

[0018] Based on the session registry, the corresponding parser is called to unpack the original data stream into key-value pairs and time tags, and convert it into a standard field set according to the semantic mapping table. The standard field set includes the park area code and the park location tag. Standardized events are generated and aggregated into a standardized event set according to the arrival order. Source and verification information are recorded, and the standardized event set is output.

[0019] Create a list of park areas, adjacency relationships between areas, and a list of entrances, exits, and access paths in the smart gateway to form a spatial topology table. Locate standardized events to areas according to park area location tags and bind them to assets. Perform consistency checks and deduplication, and write them to local storage.

[0020] Optionally, the formation of the time-series cleaning event stream specifically includes:

[0021] Receive a standardized event set, read the terminal identifier, event category, timestamp, data value, initial data quality value, location tag and source session information one by one, establish a time synchronization task according to the source session, measure the clock offset in milliseconds and the linear drift value in percentage, take the time of the first arrival of the event as the reference time, and sequentially perform offset addition, timestamp correction by multiplying by the drift value according to the reference interval, and sort in ascending order according to the corrected timestamp within each terminal;

[0022] Deduplication and debouncing are performed on the sorted event sequence of each terminal. A minimum time resolution threshold and a minimum numerical resolution threshold are set. When the time interval between adjacent events is less than the aforementioned time threshold and the numerical difference is less than the aforementioned numerical threshold, the earlier event is deleted. The duration of similar triggers is accumulated according to the set sliding window length and minimum duration. Triggers that do not reach the minimum duration are marked as suppressed.

[0023] After deduplication and jitter reduction, the event is denoised and quality assessed. Numerical data is smoothed exponentially by a set smoothing factor or filtered by median with a fixed window length to obtain smoothed data values. The time validity, numerical validity, and source and location consistency scores are calculated and summed. After comparison with the quality threshold, qualified events are marked and the data quality field is updated. The terminal identifier, event category, corrected timestamp, smoothed data value, updated data quality, location label, and source session are written into the specified key space of the local time series database to form a time-series cleaned event stream.

[0024] Optionally, the output process of the semantic event stream specifically includes:

[0025] Records belonging to the park's video sources are filtered from the time-series cleaning event stream. Continuous frame sequences are extracted according to the corrected timestamps. Target detection and threshold determination are performed on each frame in sequence to obtain detection entries containing time, temporary target number, target location and size, and confidence level. Detection entries at the same time point are written into the detection buffer. Records that are not from video sources but have image snapshots are processed in the same way, and the target detection results are output in time order.

[0026] The system performs association and tracking between adjacent time points of the target detection results. For each pair of candidate targets, it calculates three indicators: position overlap, motion continuity, and appearance fingerprint similarity. It then determines whether to establish an association according to the preset upper limit threshold. The trajectory status of the associated targets is updated, and the unassociated targets create new trajectories according to the new generation strategy. Trajectories that have not been associated for a long time are closed according to the termination strategy. The system outputs the target tracking results consisting of trajectory number, time, trajectory status, and association quality.

[0027] The target tracking results are segmented into complete segments according to the trajectory number. Based on the park's preset behavior rule library, behavior labels are determined for each segment and start and end times are given. All behavior determinations are summarized into a semantic event stream.

[0028] Optionally, obtaining the fusion event set specifically includes:

[0029] Read the spatial topology table, obtain the terminal identifier, event category and time tag of each event from the semantic event stream, and determine whether two events meet the conditions that the time interval does not exceed the time window length and the park passage distance does not exceed the radius according to the pre-configured time window length and spatial radius threshold. Event pairs that meet the conditions are written into the spatiotemporal association buffer.

[0030] Load the preset pattern sequence configuration, scan the semantic event stream in chronological order, perform pattern matching on consecutive events, and require that the event categories correspond to the pattern elements in sequence, the time tags remain non-decreasing, the time interval between adjacent events does not exceed the time tolerance, and the regions of adjacent events have reachable paths according to the topology table and the path length does not exceed the spatial tolerance. Generate pattern matching records for event subsequences that meet the conditions.

[0031] Aggregate the spatiotemporal correlation buffer and pattern matching records. Using the representative region and pattern sequence as the aggregation key, merge records that overlap in time or can be connected through spatiotemporal correlation into a single fusion event. The fusion event includes the aggregation key, the minimum and maximum start and end times, the merged event set and the supporting correlation. After completion, duplicate fusion events with the same aggregation key and intersecting time intervals are deduplicated and merged to obtain the fusion event set.

[0032] Optionally, the generation of the linkage instruction set specifically includes:

[0033] Load the park's preset grading standards and scoring parameters, receive the fusion event set, extract the duration, number of areas involved and representative path length for each fusion event, and record them as persistence, spatial influence and confidence respectively. Calculate the comprehensive score according to the non-negative constraint of the weight coefficient and the constraint that the weight sum is one. The comprehensive score is equal to persistence multiplied by the first weight plus spatial influence multiplied by the second weight plus confidence multiplied by the third weight.

[0034] The comprehensive score is segmented and judged. The first threshold, the second threshold, and the third threshold are set and arranged from low to high. When the comprehensive score is less than the first threshold, no graded warning is generated. When the comprehensive score is in the adjacent threshold interval, it is marked as the first level, the second level, or the third level, respectively. When the comprehensive score is not less than the third threshold, it is marked as the fourth level. At the same time, duplicate fusion events are deduplicated and merged within the same representative area and the same time aggregation window to generate a graded warning set.

[0035] Load the park linkage rule table into the storage medium of the pre-deployed smart gateway, retrieve the corresponding linkage template for each level warning entry of the linkage mapping table as a hierarchical warning set, and output the linkage instruction set.

[0036] Optionally, the formation of the handling status feedback specifically includes:

[0037] Analyze the action list, pre- and post-dependent dependencies and mutual exclusion relationships in the linkage instruction set, verify acyclicity and resource reachability, generate disposal batch identifiers, initialize the ready queue, running queue and completion queue, and add those without pre-dependent actions to the ready queue.

[0038] Select actions from the ready queue that do not conflict with the current running queue, assign execution terminal identifiers and parameters, generate idempotent keys and sequence numbers, set receipt timeout and maximum number of retries, batch send execution requests and start timers, and move the selected actions into the running queue.

[0039] The system continuously receives execution requests and receipts. If a request arrives successfully within the timeout period, the start time, end time, and number of attempts are recorded and the request is moved to the completion queue. If the timeout period expires, the system retryes according to the backoff strategy until the upper limit is reached. If the retry still fails, the system is marked as a failure and subsequent dependent actions are blocked. Duplicate requests and receipts are deduplicated using an idempotent key. New actions that meet the conditions are dynamically added to the ready queue. After all executable actions are completed, a handling status feedback is generated. The handling status feedback includes the handling batch identifier, the start time and end time of each action, the execution status, and the number of attempts.

[0040] Optionally, the process of integrating and reporting the evidence package specifically includes:

[0041] The system receives a set of graded warnings and feedback on the handling status. It divides the feedback on the handling status into a source fragment sequence with a fixed byte length according to the batch identifier. It records the arrival time, deadline and priority of each fragment. It establishes a sliding coding window and window step size, calculates the freshness debt of each fragment and generates a candidate set. It sets a minimum degree of freedom quota for fragments that exceed the aging threshold and declares that online fountain code asymmetric redundancy is used as the coding system for parallel reporting. The freshness debt is a measure of the urgency of each source fragment in the parallel reporting process.

[0042] The packet loss and jitter of the main link and the backup link within the estimation window are statistically analyzed. Link-aware dual-distribution degree adaptive method with online fountain code asymmetric redundancy is enabled. The degree distribution vector of each link is solved with entropy regularization optimization as the objective, and the dual distribution of the main link and the backup link is obtained. The average degree of the main link is configured to be less than that of the backup link. The main link is configured to mainly send systematically and secondarily send low degree combinations. The backup link is configured to mainly send medium and high degree combinations. The degree distribution and sending division of the two links are updated online with the link health, forming the basis of asymmetric redundancy.

[0043] To ensure the rank effectiveness of asymmetric redundancy, cross-link orthogonal pairing coding is enabled. The source fragment index is divided into several disjoint sets, and a link-set pairing task is constructed. At the same transmission time, the primary link and the backup link are forced to select different sets. The same set is prohibited from being selected repeatedly at consecutive times on the same link. An upper limit is set on the overlap of sets across time. The orthogonal pairing results of the primary link-set and the backup link-set are output.

[0044] Asymmetric redundancy is implemented during encoding and transmission. At each transmission moment, the encoded degree is extracted from the degree distribution vector of the corresponding link according to probability. The main link prioritizes extracting system fragments and a small number of low-degree combinations from the intersection of the window and the paired set. The backup link prioritizes extracting medium- and high-degree combinations accordingly. During sampling, fragments are selected without replacement from large to small based on the freshness debt. Fragments that have reached the aging threshold are supplemented with the minimum degree of freedom quota. The encoded packet is generated and the sequence number, idempotent key, window identifier, and link identifier are written into it. It is then sent to the main and backup links for parallel transmission. The receiving end aggregates and sends back the confirmation interval according to the window. The sending end only performs breakpoint resumption for unconfirmed sequence numbers and deduplicates duplicate arrivals according to the idempotent key. When the backup link has a degree of freedom gap, the average degree is increased until recovery is completed or the maximum retry threshold is reached. The records are then integrated into an evidence packet.

[0045] Optionally, the generation of the auditable record specifically includes:

[0046] Receive evidence packages, generate audit batch identifiers and archiving timestamps, extract representative areas and data categories of the park, calculate integrity summaries, and form a local primary key field set;

[0047] Archive and index according to the local primary key field set, write the link confirmation log and window statistics into the time series database and record the test point reference, register and submit the transaction log. The link confirmation log and window statistics are generated in real time based on the window-level confirmation receipt and retry process during the parallel reporting process of the primary and backup links.

[0048] Generate auditable records and append them to the local audit chain, sending them to the upper-level platform with time-series references and deduplication keys. After receiving the receipt, update the synchronization cursor and confirmation time. After synchronization is complete, set the retention period and cleanup time.

[0049] According to an embodiment of the present invention, a smart park alarm system based on a smart gateway includes the following modules:

[0050] The access and topology module is used to connect heterogeneous terminals in the park to the smart gateway and generate standardized event sets through semantic alignment, and to establish a spatial topology table of regions, adjacencies and access paths.

[0051] The cleaning and evaluation module is used to synchronize the execution time of the event set, remove duplicates, debounces, noise, and score the quality, and write qualified records into the local time-series library to form a cleaning event stream.

[0052] The semantic generation module is used to extract video frames from the clean event stream and perform target detection, correlation tracking and behavior determination, and output semantic events sorted by time.

[0053] The spatiotemporal fusion module is used to perform spatiotemporal matching and rule matching of semantic events under spatial topological constraints, and aggregate them to obtain a fused event set;

[0054] The hierarchical orchestration module is used to calculate a comprehensive score and label the level according to the park's hierarchical standards, and generate action sequences and dependency, mutual exclusion and retry parameters according to the linkage rule table.

[0055] The linkage execution module is used to issue actions, time and retry under dependency and mutual exclusion constraints, deduplicate by idempotency key and summarize the processing status feedback;

[0056] The evidence reporting module is used to integrate evidence packages and select evidence by sliding window and timely debt scheduling, perform asymmetric fountain reporting and orthogonal pairing coding, and complete archiving and audit synchronization.

[0057] The beneficial effects of this invention are:

[0058] This invention completes multi-protocol access and semantic alignment on the smart gateway side, and performs temporal cleaning and spatiotemporal fusion with spatial topology as constraints, so that the same event can achieve unified interpretation and location traceability under different manufacturers and different channels, significantly reducing false alarms and missed alarms.

[0059] This invention constructs a hierarchical evaluation and linkage orchestration link based on the park's business rules. The linkage instructions have dependency and mutual exclusion constraints, and a closed loop of receipts and retry. Execution conflicts are effectively suppressed, and the processing time and success rate can be quantified and improved.

[0060] In weak network and jitter environments, this invention adopts a link-aware dual-distribution strategy with asymmetric redundancy of online fountain codes. The main link uses a combination of systematization and low degree to ensure timeliness, while the backup link uses a combination of medium and high degree to supplement the degree of freedom. In addition, cross-link orthogonal pairing is used to improve the recovery efficiency per unit bandwidth and the probability of achieving critical fragments.

[0061] This invention achieves a closed-loop system across the entire chain from perception, analysis, handling to record keeping by parallel reporting of evidence packages, archiving of link confirmation logs and window statistics, and synchronization of auditable records of chain anchors with the platform, thereby improving the real-time performance, reliability and traceability of smart park alarms. Attached Figure Description

[0062] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0063] Figure 1 The flowchart shows a smart park alarm method and system based on a smart gateway proposed in this invention.

[0064] Figure 2 This is a schematic diagram of the algorithm structure of a smart park alarm method and system based on a smart gateway proposed in this invention. Detailed Implementation

[0065] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0066] refer to Figure 1-2 A smart park alarm method and system based on a smart gateway includes the following steps:

[0067] The multi-protocol southbound interface of the smart gateway connects to heterogeneous terminals, completes data parsing and semantic alignment, generates standardized event sets, and establishes a spatial topology table.

[0068] The standardized event set is time-synchronized, de-jittered, denoised, and quality-evaluated to form a time-series cleaned event stream and written into the local time-series database.

[0069] Target detection, tracking, and behavior recognition are performed based on a time-series cleansing event stream, and a semantic event stream is output.

[0070] Using the spatial topology table as a constraint, spatiotemporal matching, pattern matching, and event aggregation are performed on the semantic event stream to obtain a fused event set.

[0071] Based on the preset grading standards, the fusion event set is graded and evaluated, a graded warning set is generated, and a linkage instruction set is produced.

[0072] According to the linkage instruction set, the execution end is scheduled and executed under the constraints of dependency and mutual exclusion, and the status feedback is generated.

[0073] The graded warning set and the handling status feedback are integrated into an evidence package and reported in parallel through the primary and backup reporting links. The reporting process adopts online fountain code asymmetric redundancy, combined with cross-link orthogonal pairing coding to suppress rank degradation, and uses sequence number and idempotent identifier to implement confirmation, retransmission and breakpoint resume control.

[0074] Archive and index the evidence package and reported results, generate auditable records, and complete the synchronization update with the upper-level platform.

[0075] This invention significantly improves the real-time performance, accuracy, and traceability of park alarms by completing multi-protocol access, semantic alignment, and spatial topology modeling on the intelligent gateway side, followed by time-series cleaning, semantic generation, spatiotemporal fusion, hierarchical linkage, and asymmetric fountain reporting to the audit closed loop.

[0076] In this embodiment, the generation of the standardized event set and the establishment of the spatial topology table specifically include:

[0077] Configure the name, version, port and medium of the southbound protocol list on the pre-deployed smart gateway, scan and discover heterogeneous terminals within the park, complete identity authentication and address allocation, establish sessions according to terminals and protocols and register session registry, open raw data receiving queue and buffer, set packet loss alarm and rate limiting policies, and output raw data stream.

[0078] Based on the session registry, the corresponding parser is called to unpack the original data stream into key-value pairs and time tags, and convert it into a standard field set according to the semantic mapping table. The standard field set includes the park area code and the park location tag. Standardized events are generated and aggregated into a standardized event set according to the arrival order. Source and verification information are recorded, and the standardized event set is output.

[0079] Create a list of park areas, adjacency relationships between areas, and a list of entrances, exits, and access paths in the smart gateway to form a spatial topology table. Locate standardized events to areas according to park area location tags and bind them to assets. Perform consistency checks and deduplication, and write them to local storage.

[0080] This invention unifies the parsing of raw data within the gateway and generates standardized events based on a semantic mapping table. Simultaneously, it constructs a spatial topology table of campus areas and access relationships, eliminating differences in interpretation between multiple vendors and protocols, and reducing duplicate alarms and location errors.

[0081] In this embodiment, the formation of the time-series cleaning event stream specifically includes:

[0082] Receive a standardized event set, read the terminal identifier, event category, timestamp, data value, initial data quality value, location tag and source session information one by one, establish a time synchronization task according to the source session, measure the clock offset in milliseconds and the linear drift value in percentage, take the time of the first arrival of the event as the reference time, and sequentially perform offset addition, timestamp correction by multiplying by the drift value according to the reference interval, and sort in ascending order according to the corrected timestamp within each terminal;

[0083] Deduplication and debouncing are performed on the sorted event sequence of each terminal. A minimum time resolution threshold and a minimum numerical resolution threshold are set. When the time interval between adjacent events is less than the aforementioned time threshold and the numerical difference is less than the aforementioned numerical threshold, the earlier event is deleted. The duration of similar triggers is accumulated according to the set sliding window length and minimum duration. Triggers that do not reach the minimum duration are marked as suppressed.

[0084] After deduplication and jitter reduction, the event is denoised and quality assessed. Numerical data is smoothed exponentially by a set smoothing factor or filtered by median with a fixed window length to obtain smoothed data values. The time validity, numerical validity, and source and location consistency scores are calculated and summed. After comparison with the quality threshold, qualified events are marked and the data quality field is updated. The terminal identifier, event category, corrected timestamp, smoothed data value, updated data quality, location label, and source session are written into the specified key space of the local time series database to form a time-series cleaned event stream.

[0085] This invention generates a time-series cleaning event stream through time synchronization, deduplication, jitter reduction, and noise reduction, combined with quality assessment. This ensures stable and consistent input for subsequent recognition and fusion, significantly reducing false triggers caused by jitter and increasing the proportion of effective data.

[0086] In this embodiment, the output process of the semantic event stream specifically includes:

[0087] Records belonging to the park's video sources are filtered from the time-series cleaning event stream. Continuous frame sequences are extracted according to the corrected timestamps. Target detection and threshold determination are performed on each frame in sequence to obtain detection entries containing time, temporary target number, target location and size, and confidence level. Detection entries at the same time point are written into the detection buffer. Records that are not from video sources but have image snapshots are processed in the same way, and the target detection results are output in time order.

[0088] The system performs association and tracking between adjacent time points of the target detection results. For each pair of candidate targets, it calculates three indicators: position overlap, motion continuity, and appearance fingerprint similarity. It then determines whether to establish an association according to the preset upper limit threshold. The trajectory status of the associated targets is updated, and the unassociated targets create new trajectories according to the new generation strategy. Trajectories that have not been associated for a long time are closed according to the termination strategy. The system outputs the target tracking results consisting of trajectory number, time, trajectory status, and association quality.

[0089] The target tracking results are segmented into complete segments according to the trajectory number. Based on the park's preset behavior rule library, behavior labels are determined for each segment and start and end times are given. All behavior determinations are summarized into a semantic event stream.

[0090] This invention performs target detection and tracking on the video source at the edge and generates a semantic event stream based on a rule base, realizing in-situ abstraction from pixels to events, reducing backhaul bandwidth and platform-side computing pressure, and improving the timeliness of scene recognition.

[0091] In this embodiment, obtaining the fusion event set specifically includes:

[0092] Read the spatial topology table, obtain the terminal identifier, event category and time tag of each event from the semantic event stream, and determine whether two events meet the conditions that the time interval does not exceed the time window length and the park passage distance does not exceed the radius according to the pre-configured time window length and spatial radius threshold. Event pairs that meet the conditions are written into the spatiotemporal association buffer.

[0093] Load the preset pattern sequence configuration, scan the semantic event stream in chronological order, perform pattern matching on consecutive events, and require that the event categories correspond to the pattern elements in sequence, the time tags remain non-decreasing, the time interval between adjacent events does not exceed the time tolerance, and the regions of adjacent events have reachable paths according to the topology table and the path length does not exceed the spatial tolerance. Generate pattern matching records for event subsequences that meet the conditions.

[0094] Aggregate the spatiotemporal correlation buffer and pattern matching records. Using the representative region and pattern sequence as the aggregation key, merge records that overlap in time or can be connected through spatiotemporal correlation into a single fusion event. The fusion event includes the aggregation key, the minimum and maximum start and end times, the merged event set and the supporting correlation. After completion, duplicate fusion events with the same aggregation key and intersecting time intervals are deduplicated and merged to obtain the fusion event set.

[0095] This invention performs spatiotemporal matching and pattern aggregation with spatial topology as constraints, synthesizing discrete events into a fusion event set with regional and path semantics, which significantly improves the reliability of cross-regional association and complex pattern detection.

[0096] In this embodiment, the generation of the linkage instruction set specifically includes:

[0097] Load the park's preset grading standards and scoring parameters, receive the fusion event set, extract the duration, number of areas involved and representative path length for each fusion event, and record them as persistence, spatial influence and confidence respectively. Calculate the comprehensive score according to the non-negative constraint of the weight coefficient and the constraint that the weight sum is one. The comprehensive score is equal to persistence multiplied by the first weight plus spatial influence multiplied by the second weight plus confidence multiplied by the third weight.

[0098] The comprehensive score is segmented and judged. The first threshold, the second threshold, and the third threshold are set and arranged from low to high. When the comprehensive score is less than the first threshold, no graded warning is generated. When the comprehensive score is in the adjacent threshold interval, it is marked as the first level, the second level, or the third level, respectively. When the comprehensive score is not less than the third threshold, it is marked as the fourth level. At the same time, duplicate fusion events are deduplicated and merged within the same representative area and the same time aggregation window to generate a graded warning set.

[0099] Load the park linkage rule table into the storage medium of the pre-deployed smart gateway, retrieve the corresponding linkage template for each level warning entry of the linkage mapping table as a hierarchical warning set, and output the linkage instruction set.

[0100] This invention scores and classifies fusion events based on park grading standards and automatically generates linkage instructions. The actions carry dependencies, mutual exclusion constraints, and retry strategies, achieving clear handling priorities, controllable conflicts, and quantifiable execution loops.

[0101] In this embodiment, the formation of the handling status feedback specifically includes:

[0102] Analyze the action list, pre- and post-dependent dependencies and mutual exclusion relationships in the linkage instruction set, verify acyclicity and resource reachability, generate disposal batch identifiers, initialize the ready queue, running queue and completion queue, and add those without pre-dependent actions to the ready queue.

[0103] Select actions from the ready queue that do not conflict with the current running queue, assign execution terminal identifiers and parameters, generate idempotent keys and sequence numbers, set receipt timeout and maximum number of retries, batch send execution requests and start timers, and move the selected actions into the running queue.

[0104] The system continuously receives execution requests and receipts. If a request arrives successfully within the timeout period, the start time, end time, and number of attempts are recorded and the request is moved to the completion queue. If the timeout period expires, the system retryes according to the backoff strategy until the upper limit is reached. If the retry still fails, the system is marked as a failure and subsequent dependent actions are blocked. Duplicate requests and receipts are deduplicated using an idempotent key. New actions that meet the conditions are dynamically added to the ready queue. After all executable actions are completed, a handling status feedback is generated. The handling status feedback includes the handling batch identifier, the start time and end time of each action, the execution status, and the number of attempts.

[0105] This invention sends out linkage actions on the gateway side according to the queue and records the idempotent key, sequence number and acknowledgment result. It improves the success rate and predictability of linkage through timeout retry and dependency shielding mechanisms, and forms a complete handling status feedback.

[0106] In this embodiment, the process of integrating and reporting the evidence package specifically includes:

[0107] The system receives a set of graded warnings and feedback on the handling status. It divides the feedback on the handling status into a source fragment sequence with a fixed byte length according to the batch identifier. It records the arrival time, deadline and priority of each fragment. It establishes a sliding coding window and window step size, calculates the freshness debt of each fragment and generates a candidate set. It sets a minimum degree of freedom quota for fragments that exceed the aging threshold and declares that online fountain code asymmetric redundancy is used as the coding system for parallel reporting. The freshness debt is a measure of the urgency of each source fragment in the parallel reporting process.

[0108] The packet loss and jitter of the primary and backup links within the estimation window are statistically analyzed. Link-aware dual-distribution degree adaptive optimization with online fountain code asymmetric redundancy is enabled, and the degree distribution vector of each link is solved with entropy regularization optimization as the objective.

[0109] ;

[0110] in, Indicates link The degree distribution vector, Denotes the set of K-dimensional probabilistic simplex constraints. Indicates in the link Below, select degree The cost function, Indicates packet loss rate, This represents the latency jitter metric. Indicates the baseline degree distribution, Indicates the regularization weight. express Relative reference distribution The relative entropy is obtained; the dual distribution of the primary and backup links is obtained, wherein the average degree of the primary link is configured to be less than that of the backup link, the primary link is configured to mainly send systematically and supplement with low degree combinations, and the backup link is configured to mainly send medium and high degree combinations. The degree distribution and sending division of the two links are updated online with the link health, forming the basis of asymmetric redundancy.

[0111] To ensure the rank effectiveness of asymmetric redundancy, cross-link orthogonal pairing coding is enabled. The source fragment index is divided into several disjoint sets, and a link-set pairing task is constructed. At the same transmission time, the primary link and the backup link are forced to select different sets. The same set is prohibited from being selected repeatedly at consecutive times on the same link. An upper limit is set on the overlap of sets across time. The orthogonal pairing results of the primary link-set and the backup link-set are output.

[0112] Asymmetric redundancy is implemented during encoding and transmission. At each transmission moment, the encoded degree is extracted from the degree distribution vector of the corresponding link according to probability. The main link prioritizes extracting system fragments and a small number of low-degree combinations from the intersection of the window and the paired set. The backup link prioritizes extracting medium- and high-degree combinations accordingly. During sampling, fragments are selected without replacement from large to small based on the freshness debt. Fragments that have reached the aging threshold are supplemented with the minimum degree of freedom quota. The encoded packet is generated and the sequence number, idempotent key, window identifier, and link identifier are written into it. It is then sent to the main and backup links for parallel transmission. The receiving end aggregates and sends back the confirmation interval according to the window. The sending end only performs breakpoint resumption for unconfirmed sequence numbers and deduplicates duplicate arrivals according to the idempotent key. When the backup link has a degree of freedom gap, the average degree is increased until recovery is completed or the maximum retry threshold is reached. The records are then integrated into an evidence packet.

[0113] This invention employs online fountain code asymmetric redundancy combined with cross-link orthogonal pairing and sliding window time-sensitive debt scheduling. The main link systematically ensures timeliness with low degree counts, while backup links supplement degrees of freedom with medium to high degree counts, thereby improving evidence recovery rate and bandwidth efficiency in weak network conditions.

[0114] In this embodiment, the generation of auditable records specifically includes:

[0115] Receive evidence packages, generate audit batch identifiers and archiving timestamps, extract representative areas and data categories of the park, calculate integrity summaries, and form a local primary key field set;

[0116] Archive and index according to the local primary key field set, write the link confirmation log and window statistics into the time series database and record the test point reference, register and submit the transaction log. The link confirmation log and window statistics are generated in real time based on the window-level confirmation receipt and retry process during the parallel reporting process of the primary and backup links.

[0117] Generate auditable records and append them to the local audit chain, sending them to the upper-level platform with time-series references and deduplication keys. After receiving the receipt, update the synchronization cursor and confirmation time. After synchronization is complete, set the retention period and cleanup time.

[0118] This invention archives evidence packages, link confirmation logs, and window statistics in batches and establishes inverted indexes and audit chain anchors. Through incremental synchronization with the upper-level platform, it achieves verifiable full-link traceability, enhancing compliance and traceability capabilities.

[0119] A smart campus alarm system based on a smart gateway includes:

[0120] The access and topology module is used to connect heterogeneous terminals in the park to the smart gateway and generate standardized event sets through semantic alignment, and to establish a spatial topology table of regions, adjacencies and access paths.

[0121] The cleaning and evaluation module is used to synchronize the execution time of the event set, remove duplicates, debounces, noise, and score the quality, and write qualified records into the local time-series library to form a cleaning event stream.

[0122] The semantic generation module is used to extract video frames from the clean event stream and perform target detection, correlation tracking and behavior determination, and output semantic events sorted by time.

[0123] The spatiotemporal fusion module is used to perform spatiotemporal matching and rule matching of semantic events under spatial topological constraints, and aggregate them to obtain a fused event set;

[0124] The hierarchical orchestration module is used to calculate a comprehensive score and label the level according to the park's hierarchical standards, and generate action sequences and dependency, mutual exclusion and retry parameters according to the linkage rule table.

[0125] The linkage execution module is used to issue actions, time and retry under dependency and mutual exclusion constraints, deduplicate by idempotency key and summarize the processing status feedback;

[0126] The evidence reporting module is used to integrate evidence packages and select evidence by sliding window and timely debt scheduling, perform asymmetric fountain reporting and orthogonal pairing coding, and complete archiving and audit synchronization.

[0127] This invention provides a systematic modular division consisting of access and topology, cleaning and evaluation, semantic generation, spatiotemporal fusion, hierarchical orchestration, coordinated execution and evidence reporting, which facilitates engineering deployment and elastic expansion and supports integrated governance at the park level.

[0128] Example 1:

[0129] To verify the feasibility of the present invention in implementation, the present invention is applied to a typical multi-format smart park. There are office areas, dormitories and public supporting areas in the park. The front end includes a variety of heterogeneous terminals and video points. The network environment has congestion during morning and evening rush hours and individual weak coverage areas. The traditional approach mainly relies on direct connection to the platform. The device field specifications of different manufacturers are inconsistent, and video and non-video data are processed separately. Often, the same event is triggered multiple times, and it is difficult to locate the specific area path. When encountering a weak network, the evidence file upload times out, and the disposal instructions are not chained, resulting in a high false alarm rate, a low linkage success rate, and incomplete evidence collection.

[0130] The present invention deploys an intelligent gateway at the park boundary as an edge center, uniformly accesses and semantically aligns, establishes a spatial topology table composed of regions, adjacencies and entrances / exits, and completes temporal cleaning and quality assessment of video and non-video data at the edge. During on-site operation, scenarios such as abnormal gatherings at passageways, reverse pedestrian flow, and staying in sensitive areas at night can be recognized as semantic events, and spatio-temporal fusion is performed under topological constraints to obtain a set of implementable fusion events. Relying on the park grading standard, the system performs weighted scoring on the event intensity, influence range and credibility, and automatically generates a clearly graded warning and linkage instruction. The linkage covers access control, warning lights, broadcasts and camera pan-tilts, etc.

[0131] To test the reliable transmission ability in a weak network and jitter environment, the present invention fragments and schedules the evidence content on the edge side, adopts online fountain code asymmetric redundancy, the main link prioritizes systematic and low-degree combinations for timeliness, and the backup link adopts medium and high-degree combinations to supplement the degrees of freedom; the two links avoid coverage overlap through cross-link orthogonal pairing, and combine sliding window and freshness debt priority strategies to ensure that approaching deadlines and aging fragments are preferentially selected and have the smallest quota.

[0132] In actual measurement, taking the same batch of cross-region staying events as an example, when the link packet loss rate increases during the commuting peak period, the evidence complete recovery rate decreases significantly when the main link reports alone, while the main-backup asymmetric parallel reporting can still complete the recovery within the window, and gives link confirmation logs and window statistics, and then uniformly archives them to object and temporal storage, automatically generates an auditable record and performs incremental synchronization with the upper-level platform. To make the comparison clearly readable, the following gives the comparison of core indicators, as shown in Table 1:

[0133] Table 1 Comparative Evaluation Table of Smart Park Alarm Systems

[0134] Indicator Item direct connection to traditional platforms This invention Event false alarm rate 6.8% 2.1% Event detection rate 91.5% 97.3% Processing time across the entire chain 18.4s 9.7s Linkage success rate 92.0% 98.2% Evidence completeness recovery rate 83.6% 96.8% Uplink bandwidth usage 1.00 0.78 Audit traceability 58% 100%

[0135] As shown in Table 1, this invention significantly outperforms the baseline in several key indicators: the false alarm rate is reduced from 6.8% to 2.1%, and the detection rate is increased from 91.5% to 97.3%; the median processing time across the entire link is shortened from 18.4s to 9.7s, and the linkage success rate is increased from 92.0% to 98.2%. In weak network scenarios (packet loss rate ≈ 15%), the evidence integrity recovery rate is increased from 83.6% to 96.8%, and the uplink bandwidth usage is reduced to 0.78 at the same integrity level; audit traceability coverage is achieved from 58% to 100%.

[0136] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart park alarm method based on an intelligent gateway, characterized in that, Includes the following steps: The multi-protocol southbound interface of the smart gateway connects to heterogeneous terminals, completes data parsing and semantic alignment, generates standardized event sets, and establishes a spatial topology table. The standardized event set is time-synchronized, de-jittered, denoised, and quality-evaluated to form a time-series cleaned event stream and written into the local time-series database. Target detection, tracking, and behavior recognition are performed based on a time-series cleansing event stream, and a semantic event stream is output. Using the spatial topology table as a constraint, spatiotemporal matching, pattern matching, and event aggregation are performed on the semantic event stream to obtain a fused event set. Based on the preset grading standards, the fusion event set is graded and evaluated, a graded warning set is generated, and a linkage instruction set is produced. According to the linkage instruction set, the execution end is scheduled and executed under the constraints of dependency and mutual exclusion, and the status feedback is generated. The graded warning set and the handling status feedback are integrated into an evidence package and reported in parallel through the primary and backup reporting links. The reporting process adopts online fountain code asymmetric redundancy and combines cross-link orthogonal pairing coding to suppress rank degradation. Archive and index the evidence package and reported results, generate auditable records, and complete the synchronization update with the upper-level platform.

2. The smart park alarm method based on a smart gateway according to claim 1, characterized in that, The generation of the standardized event set and the establishment of the spatial topology table specifically include: Configure the name, version, port and medium of the southbound protocol list on the smart gateway, scan and discover heterogeneous terminals within the park, complete identity authentication and address allocation, establish sessions according to terminals and protocols and register them in the session registry, and output the raw data stream. Based on the session registry, the corresponding parser is called to unpack the raw data stream into key-value pairs and timestamps, convert it into a set of standard fields according to the semantic mapping table, generate standardized events and aggregate them into a standardized event set in the order of arrival, record the source and verification information, and output the standardized event set; Create a list of park areas, adjacency relationships between areas, and a list of entrances, exits, and access paths in the smart gateway to form a spatial topology table.

3. The smart park alarm method based on a smart gateway according to claim 1, characterized in that, The formation of the time-series cleaning event stream specifically includes: Receive standardized event sets, establish time synchronization tasks according to the source session, measure clock offset and linear drift value, correct timestamps, and sort them in ascending order of the corrected timestamps within each terminal; Perform deduplication and debouncing on the sorted event sequence of each terminal; After deduplication and dejittering, the events are denoised and quality evaluated to obtain smoothed data values. The scores for time validity, numerical validity, and consistency of source and location are calculated and summed. After comparison with the quality threshold, qualified events are marked and the data quality field is updated to form a time-series cleaned event stream.

4. The smart park alarm method based on a smart gateway according to claim 1, characterized in that, The output process of the semantic event stream specifically includes: Records belonging to the park's video sources are filtered out from the time-series cleaning event stream. Continuous frame sequences are extracted according to the corrected timestamps. Target detection and threshold determination are performed on each frame in sequence to obtain detection entries. Detection entries at the same time point are written into the detection buffer, and the target detection results are output in time order. Perform correlation and tracking between adjacent time points of the target detection results, and output the target tracking results consisting of trajectory numbers; The target tracking results are segmented into complete segments according to the trajectory number. Based on the park's preset behavior rule library, behavior labels are determined for each segment and start and end times are given. All behavior determinations are summarized into a semantic event stream.

5. The smart park alarm method based on a smart gateway according to claim 1, characterized in that, The specific process of obtaining the fusion event set includes: Read the spatial topology table, obtain the terminal identifier, event category and time tag of each event from the semantic event stream, and determine whether two events meet the conditions that the time interval does not exceed the time window length and the park passage distance does not exceed the radius according to the pre-configured time window length and spatial radius threshold. Event pairs that meet the conditions are written into the spatiotemporal association buffer. Load the preset pattern sequence configuration, scan the semantic event stream in chronological order, perform pattern matching on continuous events, and generate pattern matching records for event subsequences that meet the conditions; Aggregate the spatiotemporal correlation buffer and pattern matching records, using the representative region and pattern sequence as the aggregation key. Merge records that overlap in time or can be connected through spatiotemporal correlation into a single fusion event. After completion, perform deduplication and merging on duplicate fusion events with the same aggregation key and intersecting time intervals to obtain a fusion event set.

6. The smart park alarm method based on a smart gateway according to claim 1, characterized in that, The generation of the linkage instruction set specifically includes: Load the park's preset grading standards and scoring parameters, receive the fusion event set, extract the duration, number of areas involved and representative path length for each fusion event, and record them as persistence, spatial impact and confidence, respectively, and calculate the comprehensive score. The comprehensive score is segmented and judged. The first threshold, the second threshold, and the third threshold are set and arranged from low to high. At the same time, duplicate fusion events are deduplicated and merged within the same representative region and the same time aggregation window to generate a graded warning set. Load the park linkage rule table into the storage medium of the pre-deployed smart gateway, retrieve the corresponding linkage template for each level warning entry of the linkage mapping table as a hierarchical warning set, and output the linkage instruction set.

7. The smart park alarm method based on a smart gateway according to claim 1, characterized in that, The formation of the handling status feedback specifically includes: Analyze the action list, pre- and post-dependent dependencies and mutual exclusion relationships in the linkage instruction set, verify acyclicity and resource reachability, generate disposal batch identifiers, initialize the ready queue, running queue and completion queue, and add those without pre-dependent actions to the ready queue. Select actions from the ready queue that do not conflict with the current running queue, assign execution terminal identifiers and parameters, generate idempotent keys and sequence numbers, set receipt timeout and maximum number of retries, batch send execution requests and start timers, and move the selected actions into the running queue. It continuously receives execution requests and receipts, removes duplicate requests and receipts using an idempotent key, dynamically adds new actions that meet the conditions to the ready queue, and generates a processing status feedback after all executable actions are completed.

8. The smart park alarm method based on a smart gateway according to claim 1, characterized in that, The process of integrating and reporting the evidence package specifically includes: Receive the graded warning set and the handling status feedback, divide the handling status feedback into source fragment sequences with a fixed byte length according to the handling batch identifier, establish a sliding coding window and window step size, calculate the freshness debt of each fragment and generate a candidate set, and set a minimum degree of freedom quota for fragments that exceed the aging threshold. The packet loss and jitter of the main link and backup link within the estimation window are statistically analyzed. The link-aware dual-distribution degree adaptive method with online fountain code asymmetric redundancy is enabled. The degree distribution vector of each link is solved with entropy regularization optimization as the goal, and the dual distribution of the main and backup links is obtained. The degree distribution and transmission division of the two links are updated online with the link health, forming the basis of asymmetric redundancy. Enable cross-link orthogonal pairing coding, divide the source fragment index into several disjoint sets, construct the link-set pairing task, force the primary link and backup link to select different sets at the same transmission time, and output the orthogonal pairing results of the primary link-set and backup link-set; During encoding and transmission, asymmetric redundancy is implemented. At each transmission moment, the encoding degree is extracted from the degree distribution vector of the corresponding link according to probability. During sampling, fragments are selected without replacement according to the freshness of the debt from large to small. Encoded packets are generated and the sequence number, idempotent key, window identifier and link identifier are written. The packets are then sent to the primary and backup links in parallel. The sending end only performs breakpoint resume transmission for unconfirmed sequence numbers and deduplicates duplicate arrivals according to the idempotent key. The records are then integrated into evidence packets.

9. A smart park alarm method based on a smart gateway according to claim 1, characterized in that, The generation of the auditable records specifically includes: Receive evidence packages, generate audit batch identifiers and archiving timestamps, extract representative areas and data categories of the park, calculate integrity summaries, and form a local primary key field set; Archive and index according to the local primary key field set, write the link confirmation log and window statistics into the time series database and record the measurement point reference, register and commit the transaction log; Generate auditable records and append them to the local audit chain, sending them to the upper-level platform with time-series references and deduplication keys. After receiving the receipt, update the synchronization cursor and confirmation time. After synchronization is complete, set the retention period and cleanup time.

10. A smart park alarm system based on a smart gateway, comprising executing the smart park alarm method based on a smart gateway as described in any one of claims 1 to 9, characterized in that, include: The access and topology module is used to connect heterogeneous terminals in the smart gateway to the campus and generate standardized event sets through semantic alignment, and to establish a spatial topology table of regions, adjacencies and access paths. The cleaning and evaluation module is used to synchronize the execution time of the event set, remove duplicates, debounces, noise, and score the quality, and write qualified records into the local time-series library to form a cleaning event stream. The semantic generation module is used to extract video frames from the clean event stream and perform target detection, correlation tracking and behavior determination, and output semantic events sorted by time. The spatiotemporal fusion module is used to perform spatiotemporal matching and rule matching of semantic events under spatial topological constraints, and aggregate them to obtain a fused event set; The hierarchical orchestration module is used to calculate a comprehensive score and label the level according to the park's hierarchical standards, and generate action sequences and dependency, mutual exclusion and retry parameters according to the linkage rule table. The linkage execution module is used to issue actions, time and retry under dependency and mutual exclusion constraints, deduplicate by idempotency key and summarize the processing status feedback; The evidence reporting module is used to integrate evidence packages and select evidence by sliding window and timely debt scheduling, perform asymmetric fountain reporting and orthogonal pairing coding, and complete archiving and audit synchronization.

Images