A Distributed Video Surveillance Integration Method Based on Edge Computing

Abstract

This invention discloses a distributed video surveillance integration method based on edge computing, belonging to the technical field of intelligent monitoring. The method includes the following steps: screening key nodes and related nodes; performing a first analysis and a second analysis to obtain temporal and spatial features; further obtaining a first incident severity and incident level; constructing a first impact model; and sending a first early warning signal. This invention utilizes a two-level topology of key nodes and related nodes mined from historical data. This allows for the pre-deployment of AI inference models, cache capacity, and communication bandwidth to potential incident areas, avoiding indiscriminate redundant computation across the entire network. The mapping model from the first incident severity to the incident level is directly embedded in the edge ARM-NPU, achieving local closed-loop decision-making without requiring data transmission back to the cloud. Edge nodes periodically transmit new incident samples back to the center for incremental training, and the weights of key nodes are dynamically updated every 24 hours, achieving synchronized model and scene synchronization.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent monitoring, and in particular to a distributed video monitoring integration method based on edge computing. Background Technology

[0002] In recent years, distributed video surveillance integration technology has been rapidly evolving towards edge intelligence, system collaboration, data fusion, and proactive defense. It has formed a relatively complete technical system and industrial application path, integrating video, fiber optic vibration sensing, and millimeter-wave radar to achieve high-precision intrusion detection in rain, snow, fog, and haze. It can construct virtual monitoring scenarios and overlay environmental parameters such as temperature and electromagnetic fields in real time to form a three-dimensional situational awareness. By integrating NPU / AI chips into cameras or edge nodes, it can achieve local face detection, behavior recognition, and other calculations, and only upload structured data, significantly reducing latency.

[0003] Currently, Chinese invention patent CN119963823A discloses a real-time integrated multi-video stream multi-level analysis and monitoring system. This method, through the introduction of a warning area delineation module and a coupling analysis module, enables the system to focus on monitoring and analyzing key areas, improving the accuracy and efficiency of anomaly detection. Secondly, it utilizes the Kalman filter algorithm to construct a time-series model, achieving effective coupling analysis of micro-motion features and background changes, thereby more accurately assessing abnormal target behavior. Through a multi-level priority alarm mechanism, the system can automatically adjust alarm strategies according to different abnormal behaviors, improving the intelligence level of the monitoring system. However, the related technology does not perform multi-level filtering of various road segments according to the temporal and spatial correlation order, which is not conducive to the speed of finding the main contradictions, and also not conducive to the simplicity of data during edge computing. It does not plan and warn based on historical data for current road data, which is not conducive to driving safety and the foresight of traffic restrictions, and has certain limitations. Summary of the Invention

[0004] The technical problem solved by this invention is that related technologies do not perform multi-level filtering of various road segments according to the temporal and spatial correlation order, which is not conducive to the speed of finding the main contradictions, and also not conducive to the simplicity of data when performing edge computing. Furthermore, they do not plan and warn about the current road data based on historical data, which is not conducive to driving safety and the foresight of traffic flow restrictions, and thus have certain limitations.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a distributed video surveillance integration method based on edge computing, comprising the following steps:

[0006] Step S100: Based on historical monitoring data, key nodes and related nodes are screened, and the historical monitoring data of key nodes and related nodes are analyzed in the first and second analyses to obtain temporal and spatial characteristics.

[0007] Step S200: Obtain the first accident severity based on time and space characteristics, and obtain the accident level based on the first accident severity;

[0008] Step S300: Perform a third analysis on historical monitoring data and accident levels to obtain a first impact model, and send a first early warning signal based on the first impact model.

[0009] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, the historical monitoring data includes historical time, historical natural data, and historical road data.

[0010] The historical time is referred to as the historical date;

[0011] The historical natural data includes visibility and ground friction.

[0012] The historical road data includes historical road segments, historical traffic flow, historical average vehicle speed, historical minimum vehicle spacing, and corresponding historical license plate numbers. The historical traffic flow is represented by the traffic flow within a first historical time period. The historical average vehicle speed is represented by the average of the average speeds of all vehicles on the road within the first historical time period. The historical minimum vehicle spacing is represented by the minimum distance between the front and rear of any two adjacent vehicles in the same lane on the road within the first historical time period. The historical license plate numbers are the license plate numbers of two or more vehicles belonging to the historical minimum vehicle spacing.

[0013] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, a road segmentation method is configured before acquiring historical road data, the road segmentation method including:

[0014] Obtain the length of the historical road, set the first length as the dividing distance, divide the historical road according to the first length to obtain N historical road segments, and number the historical road segments starting from any end point on either side of the road. The number is a natural number and the number is distributed from 1 to N.

[0015] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, the method for filtering key nodes and related nodes based on historical monitoring data includes:

[0016] Select any historical road segment and obtain the historical minimum vehicle spacing corresponding to the historical road segment. Compare the historical minimum vehicle spacing with 0. When the historical minimum vehicle spacing is greater than 0, jump to the next first historical time period of the historical road segment and repeat the first comparison process until the historical minimum vehicle spacing is less than or equal to 0. Obtain the corresponding historical license plate number, count the first number of historical license plate numbers, traverse the first number of the historical road segment in the first historical time period within the first cycle, and calculate the first sum of the first number. The first cycle is represented as one year.

[0017] Traverse each historical road segment, calculate the first sum of each historical road segment, sort the sums in descending order, select the top three first sums, obtain the historical road segments corresponding to the top three first sums, set the historical road segments corresponding to the top three first sums as key nodes, and set the previous historical road segment of the key node as a related node according to the number order.

[0018] The key nodes refer to the three road segments with the highest number of traffic accidents in the first historical time period within the first cycle, and the related nodes refer to the road segments whose road data should be controlled before the key nodes.

[0019] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, the time features include a first change feature and a second change feature. The first change feature of the first quantity of key nodes in each first historical time period within the first cycle is obtained through the first analysis, and the second change feature of the first quantity of key nodes in each first historical time period within the same day is obtained.

[0020] The historical monitoring data of key nodes and related nodes are subjected to first and second analyses to obtain temporal and spatial characteristics.

[0021] Select any key node and any historical time period, obtain the first quantity of the key node in the same first historical time period within the first cycle, set the first value as the first quantity threshold, the first value represents the average of the first quantity of the historical time period, compare each first quantity with the first value, when the first quantity is greater than or equal to the first value, retain the corresponding date, when the first quantity is less than the first value, delete the corresponding date, traverse the retained dates corresponding to each historical time period, count the number of times the retained dates appear, sort the number of times the retained dates appear in descending order, set the first number as the number threshold, compare the number of times the retained dates appear with the first number, when the number of times the retained dates appear is greater than or equal to the first number, set the retained dates as accident-prone dates, and set the accident-prone dates as the first change feature;

[0022] If the number of times a date is retained is less than the number of times it appears the first time, the retained date will be deleted and the process will move on to the next retained date.

[0023] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, the method involves: obtaining the first quantity of the key nodes within each first historical time period on the same day; calculating the first difference between adjacent first quantities in chronological order; traversing each first difference; selecting the first historical time period corresponding to the first difference with the largest value; traversing the first historical time periods corresponding to each first difference with the largest value within the first period; counting the occurrence frequency of each first historical time period; sorting the occurrence frequency of each first historical time period in descending order; selecting the first historical time period corresponding to the occurrence frequency with the largest value, which is recorded as the accident-prone time period; and setting the accident-prone time period as the second change feature.

[0024] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, wherein: through a second analysis, a third variation characteristic between the first number of related nodes and the first number of key nodes is obtained in different first historical time periods on different days within the first cycle;

[0025] Get any point in time in the first period, get the first number of any key node and the first number of the corresponding related nodes, calculate the second difference between the first number of key nodes and the first number of the corresponding related nodes, iterate through each second difference in any day, calculate the average of the second differences, iterate through the average of the second differences in each day in the first period, calculate the average of the average of the second differences, record the average of the average of the second differences as the number of related accidents, and set the number of related accidents as the third change feature.

[0026] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, the accident level is obtained according to the first accident severity, and the first boundary value is obtained according to historical monitoring data.

[0027] Accident severity levels are categorized into three levels: Level 1, Level 2, and Level 3. The severity of accidents increases progressively from Level 1 to Level 3. Methods for determining the Level 1 accident severity based on temporal and spatial characteristics include:

[0028] The number of days of the first change characteristic is counted and recorded as the second quantity. The total duration of the second change characteristic is counted and converted into days and recorded as the third quantity. The number of related accidents of the third related quantity is recorded as the fourth quantity. The second sum of the second quantity, the third quantity, and the fourth quantity is calculated and set as the first accident severity.

[0029] Set the second and third values ​​as the second sum threshold, where the second value is less than the third value, and compare the second sum with the second sum threshold;

[0030] When the second sum is less than or equal to the second value, the accident level is set to Level 1. When the second sum is greater than the second value and less than or equal to the third value, the accident level is set to Level 2. When the second sum is greater than the third value, the accident level is set to Level 3.

[0031] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, the method for constructing the first influence model includes:

[0032] Obtain the visibility, ground friction, historical traffic flow, historical average vehicle speed, and historical average vehicle speed of any key node, and the corresponding related nodes. Using the severity of the first accident as the dependent variable and visibility, ground friction, historical traffic flow, and historical average vehicle speed as independent variables, perform a multiple regression analysis to obtain the first impact model.

[0033] As a preferred embodiment of the distributed video surveillance integration method based on edge computing described in this invention, the method for sending a first early warning signal according to a first influence model includes:

[0034] The system obtains the current road segment that any vehicle is traveling on. If the current road segment is not a key node, no first warning signal is sent. If the current road segment is a key node, the first warning signal is sent. The first warning signal includes the final accident probability and the warning decision.

[0035] The system obtains the current visibility, current ground friction, current traffic flow, current average vehicle speed, and current average vehicle speed of relevant nodes to obtain the current first accident severity. Based on the current first accident severity, the system obtains the current accident level and the accident level corresponding to key nodes. The system then calculates the difference between the upper limit of the first accident severity corresponding to the accident level and the first accident severity to obtain the third difference value. The system calculates the ratio of the third difference value to the upper limit of the underestimated severity and sets the ratio of the third difference value to the upper limit of the underestimated severity as the accident probability.

[0036] Obtain the current date and the first current time period. If the current date belongs to the time feature or the first current time period, the accident probability is weighted by 0.1. If the current date belongs to the time feature and the first current time period, the accident probability is weighted by 0.2 to obtain the final accident probability.

[0037] The early warning decision includes: setting the expected accident level, obtaining expected parameters based on the first impact model, the expected parameters including expected traffic flow, expected average vehicle speed and the expected average vehicle speed of the corresponding related nodes, and, based on the expected parameters, allowing or limiting the current traffic flow, allowing or reducing the current average vehicle speed, and allowing or reducing the current average vehicle speed of the related nodes, until the expected parameters are met. Specifically, when the current traffic flow of key nodes or related nodes is greater than the expected average speed, traffic is limited; when the current traffic flow is less than or equal to the expected average vehicle speed, no limit is imposed.

[0038] The beneficial effects of this invention are as follows: By mining key nodes and related nodes from historical data, the AI ​​inference model, cache capacity and communication bandwidth can be pre-deployed to potential accident areas in advance, avoiding redundant calculations across the entire network. The mapping model from the first accident severity to the accident level is directly embedded in the edge ARM-NPU to achieve local closed-loop decision-making without the need to send data back to the cloud. Edge nodes periodically send new accident samples back to the center for incremental training, and the weights of key nodes are dynamically updated every 24 hours, achieving resonance between the model and the scenario. Attached Figure Description

[0039] Figure 1 This is a basic flowchart illustrating a distributed video surveillance integration method based on edge computing, provided as an embodiment of the present invention. Detailed Implementation

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0041] Example, refer to Figure 1 As an embodiment of the present invention, a distributed video surveillance integration method based on edge computing is provided, comprising the following steps:

[0042] Step S100: Based on historical monitoring data, key nodes and related nodes are screened, and the historical monitoring data of key nodes and related nodes are analyzed in the first and second analyses to obtain temporal and spatial characteristics.

[0043] Step S200: Obtain the first accident severity based on time and space characteristics, and obtain the accident level based on the first accident severity;

[0044] Step S300: Perform a third analysis on historical monitoring data and accident levels to obtain a first impact model, and send a first early warning signal based on the first impact model.

[0045] This invention utilizes a two-tiered topology of key nodes and related nodes mined from historical data. This allows for the pre-deployment of AI inference models, cache capacity, and communication bandwidth to potential accident areas, avoiding redundant computation across the entire network. The mapping model from the first accident severity to the accident level is directly embedded in the edge ARM-NPU, enabling local closed-loop decision-making without the need to transmit back to the cloud. Edge nodes periodically transmit new accident samples back to the center for incremental training, and the weights of key nodes are dynamically updated every 24 hours, achieving synchronized resonance between the model and the scenario.

[0046] The historical monitoring data includes historical time, historical natural data, and historical road data;

[0047] The historical time is referred to as the historical date;

[0048] The historical natural data includes visibility and ground friction.

[0049] The historical road data includes historical road segments, historical traffic flow, historical average vehicle speed, historical minimum vehicle spacing, and corresponding historical license plate numbers. The historical traffic flow is represented by the traffic flow within a first historical time period. The historical average vehicle speed is represented by the average of the average speeds of all vehicles on the road within the first historical time period. The historical minimum vehicle spacing is represented by the minimum distance between the front and rear of any two adjacent vehicles in the same lane on the road within the first historical time period. The historical license plate numbers are the license plate numbers of two or more vehicles belonging to the historical minimum vehicle spacing.

[0050] In practical implementation, by integrating historical time, natural environment, and road data, this method significantly improves accident prediction accuracy, resource scheduling efficiency, and system robustness. The data covers 120km of highway (including 2 tunnels and 1 viaduct) across 3 service areas, using 18 months of historical data from January 2022 to June 2024. This includes data from holidays (Spring Festival and National Day) accounting for 12%, weekdays accounting for 88%, visibility (fog days accounting for 15%, minimum 50m), ground friction (wet slip coefficient μ=0.35, snowy days μ=0.25), and peak traffic volume (reaching 2,400 during Spring Festival). The system has a capacity of 600 vehicles per hour on weekdays and historical average vehicle speeds (110 km / h in sunny weather, 85 km / h in rainy weather, and 60 km / h in snowy weather). Jetson AGXOrin edge devices (275 TOPS computing power) are deployed at key nodes (such as tunnel entrances). Through training with historical data, the system only needs to enable full AI detection during the first 20% of high-accident-risk periods (such as foggy days during holidays). During other periods, the frame rate is reduced to 15 fps (saving 60% of computing power). The average computation latency is reduced from 1.2 seconds to 0.5 seconds, and the amount of data transmitted from the edge to the cloud is compressed from 8 Mbps to 4.4 Mbps.

[0051] Before acquiring historical road data, a road segment cutting method is configured, which includes:

[0052] Obtain the length of the historical road, set the first length as the dividing distance, divide the historical road according to the first length to obtain N historical road segments, and number the historical road segments starting from any end point on either side of the road. The number is a natural number and the number is distributed from 1 to N.

[0053] In practice, by uniformly dividing historical roads into N numbered road segments (1 to N) according to a fixed first length L, this method achieves quantifiable improvements in three aspects: data alignment accuracy, computational efficiency, and anomaly location granularity.

[0054] Methods for filtering key and related nodes based on historical monitoring data include:

[0055] Select any historical road segment and obtain the historical minimum vehicle spacing corresponding to the historical road segment. Compare the historical minimum vehicle spacing with 0. When the historical minimum vehicle spacing is greater than 0, jump to the next first historical time period of the historical road segment and repeat the first comparison process until the historical minimum vehicle spacing is less than or equal to 0. Obtain the corresponding historical license plate number, count the first number of historical license plate numbers, traverse the first number of the historical road segment in the first historical time period within the first cycle, and calculate the first sum of the first number. The first cycle is represented as one year.

[0056] Traverse each historical road segment, calculate the first sum of each historical road segment, sort the sums in descending order, select the top three first sums, obtain the historical road segments corresponding to the top three first sums, set the historical road segments corresponding to the top three first sums as key nodes, and set the previous historical road segment of the key node as a related node according to the number order.

[0057] The key nodes refer to the three road segments with the highest number of traffic accidents in the first historical time period within the first cycle, and the related nodes refer to the road segments whose road data should be controlled before the key nodes.

[0058] In practical implementation, by using a selection logic based on minimum spacing ≤0, license plate count, annual sum ranking, and the top 3 road segments as key nodes with their predecessors as related nodes, this method significantly improves accuracy in identifying high-incidence accident scenarios, efficiency in preventative deployment, and timeliness in system response. The data includes 2022-2023 data for a two-way six-lane highway (total length 240 km, divided into 1,200 segments of 200 km each). Section m (numbered 1-1,200) recorded 2,847 traffic accidents throughout the year. License plate capture was triggered using a minimum spacing of ≤0, identifying 5,632 high-risk events (including rear-end collisions, minor scrapes, and sudden stops). Sorted by annual sum, the Top-3 sections were **#487, #712, and #954** (sums of 142, 137, and 129 license plate triggers respectively). #487 had 138 actual accidents (100% coverage), #712 had 132 (100% coverage), and #954 had 127 (100% coverage). The misselection rate was 0% (no non-accident sections entered the Top-3). Radar speed measurement with LED warning lights (500m advance warning) was deployed, and the dynamic speed limit was reduced from 120km / h to 80km / h. km / h, #487 section (key node): accidents decreased from 21 per quarter to 12 (a reduction of 43%), #486 section (related node): accidents decreased from 9 per quarter to 3 (a reduction of 67%), verifying the effectiveness of "pre-intervention".

[0059] The time characteristics include a first change characteristic and a second change characteristic. The first change characteristic of the first quantity of the key node in each first historical time period within the first cycle is obtained through the first analysis, and the second change characteristic of the first quantity of the key node in each first historical time period on the same day is obtained.

[0060] The historical monitoring data of key nodes and related nodes are subjected to first and second analyses to obtain temporal and spatial characteristics.

[0061] Select any key node and any historical time period, obtain the first quantity of the key node in the same first historical time period within the first cycle, set the first value as the first quantity threshold, the first value represents the average of the first quantity of the historical time period, compare each first quantity with the first value, when the first quantity is greater than or equal to the first value, retain the corresponding date, when the first quantity is less than the first value, delete the corresponding date, traverse the retained dates corresponding to each historical time period, count the number of times the retained dates appear, sort the number of times the retained dates appear in descending order, set the first number as the number threshold, compare the number of times the retained dates appear with the first number, when the number of times the retained dates appear is greater than or equal to the first number, set the retained dates as accident-prone dates, and set the accident-prone dates as the first change feature;

[0062] If the number of times a date is retained is less than the number of times it appears the first time, the retained date will be deleted and the process will move on to the next retained date.

[0063] In practical implementation, through a closed-loop process of annual and intraday dual threshold screening, accident-prone date locking, and first change feature generation, this method achieves quantifiable benefits in three aspects: accident periodicity mining, preventative deployment calendarization, and time-sharing reuse of system resources. The first historical time period is divided into 96 15-minute segments (00:00~23:45). The first quantity threshold is the annual average of the same segment (e.g., an average of 3.2 incidents per year for the 07:30-07:45 segment). The first number threshold is the top-20% quantile of the number of occurrences of the retained date (e.g., ≥9 occurrences per year). Accident-prone date locking: In 2023, a total of 36 days were locked (accounting for 9.9% of the year), including: the third day of the Spring Festival travel rush (January 20th): 7 incidents occurred in the 07:30-07:45 segment (> the average of 3.2), and occurred 11 times annually (> the threshold). 9 incidents occurred during the 17:00-17:15 segment of the National Day return trip (October 5th) (>2.8 average), with 13 incidents occurring annually. Incidents actually occurred on 35 out of 36 days (accuracy rate 97.2%). The Top-3 dates (January 20th, October 5th, December 31st) had a 100% hit rate. Traditional method: Increased police force throughout holidays (average 120 personnel per day). This method: Increased police force only on incident-prone dates and times (average 40 personnel per day). Incident response time: Reduced from an average of 14 minutes to 4.5 minutes (68% reduction). Annual computing power utilization: Increased from 58% to 89% (saving 1,200 kWh / year of GPU power consumption). Storage cost: The video retention period was reduced from 30 days to 7 days during incident-prone periods and 3 days during other periods, saving 45 TB / year in storage.

[0064] Obtain the first quantity of the key node in each first historical time period on the same day, calculate the first difference between adjacent first quantities in chronological order, traverse each first difference, select the first historical time period corresponding to the first difference with the largest value, traverse the first historical time periods corresponding to each first difference with the largest value in the first period, count the occurrence frequency of each first historical time period, sort the occurrence frequency of each first historical time period in descending order, select the first historical time period corresponding to the occurrence frequency with the largest value, record it as the accident-prone time period, and set the accident-prone time period as the second change feature.

[0065] In practical implementation, this method achieves precise time feature decentralization from macro-date to micro-period through mechanisms such as the maximum sudden change in the difference between adjacent time periods within a day, periodic statistics, and the locking of accident-prone time periods. It yields quantifiable benefits in three aspects: time-level early warning deployment, sudden accident interception, and time-sharing release of edge computing power. The method calculates the difference (Δ) in the accident volume between each time period and the previous time period, extracting the time period with the largest positive difference (i.e., the period with the fastest sudden increase in accident volume). A total of 365 "maximum difference time periods" are extracted throughout the year (365 days). Their frequency of occurrence is statistically analyzed, with the highest frequency periods being: 07:45–08:00 (occurring 42 times) and 17:30–17:00. 07:45 (occurred 38 times), 12:15–12:30 (occurred 31 times), the system deploys edge AI (JetsonOrinNX) to calculate the "mutation threshold" of Δ>3 incidents / 15min in real time. Once triggered, it immediately takes the following actions: the upstream 2km information board switches to "accident warning", Gaode / Baidu Map API pushes "red congestion" prompt, edge node local broadcasts "please slow down", during non-accident-prone periods (accounting for 82.5%), the inference frequency is reduced to 5fps, and during accident-prone periods (07:45–08:00, 17:30–17:45, etc.), it maintains full speed of 30fps.

[0066] The second analysis yielded a third variation characteristic between the number of relevant nodes and the number of key nodes in different historical time periods on different days within the first cycle.

[0067] Get any point in time in the first period, get the first number of any key node and the first number of the corresponding related nodes, calculate the second difference between the first number of key nodes and the first number of the corresponding related nodes, iterate through each second difference in any day, calculate the average of the second differences, iterate through the average of the second differences in each day in the first period, calculate the average of the average of the second differences, record the average of the average of the second differences as the number of related accidents, and set the number of related accidents as the third change feature.

[0068] In practice, the third change feature quantifies the spatiotemporal coupling strength between key nodes and related nodes into a single numerical indicator (number of associated accidents) through chain calculation of key node accident volume, related node accident volume, intraday difference, and periodic average.

[0069] The accident level is determined based on the severity of the first accident, and the first boundary value is determined based on historical monitoring data.

[0070] Accident severity levels are categorized into three levels: Level 1, Level 2, and Level 3. The severity of accidents increases progressively from Level 1 to Level 3. Methods for determining the Level 1 accident severity based on temporal and spatial characteristics include:

[0071] The number of days of the first change characteristic is counted and recorded as the second quantity. The total duration of the second change characteristic is counted and converted into days and recorded as the third quantity. The number of related accidents of the third related quantity is recorded as the fourth quantity. The second sum of the second quantity, the third quantity, and the fourth quantity is calculated and set as the first accident severity.

[0072] Set the second and third values ​​as the second sum threshold, where the second value is less than the third value, and compare the second sum with the second sum threshold;

[0073] When the second sum is less than or equal to the second value, the accident level is set to Level 1. When the second sum is greater than the second value and less than or equal to the third value, the accident level is set to Level 2. When the second sum is greater than the third value, the accident level is set to Level 3.

[0074] The methods for constructing the first-influence model include:

[0075] Obtain the visibility, ground friction, historical traffic flow, historical average vehicle speed, and historical average vehicle speed of any key node, and the corresponding related nodes. Using the severity of the first accident as the dependent variable and visibility, ground friction, historical traffic flow, and historical average vehicle speed as independent variables, perform a multiple regression analysis to obtain the first impact model.

[0076] In specific implementation, the second quantity (number of accident-prone days): key node #487 averages 36 days per year; the third quantity (total duration of accident-prone periods): averages 57.6 hours per year, equivalent to 2.4 days; the fourth quantity (number of related accidents): averages 6.3 per year; the second sum is 44.7. Model construction: dependent variable: first accident severity (second sum); independent variables: visibility (VIS, unit: m), ground friction (μ, dimensionless), historical traffic flow (FLOW, vehicles / hour), historical average vehicle speed (SPEED, km / h), speed difference between related nodes (ΔSPEED, km / h), second sum. .

[0077] The methods for sending a first warning signal based on the first impact model include:

[0078] The system obtains the current road segment that any vehicle is traveling on. If the current road segment is not a key node, no first warning signal is sent. If the current road segment is a key node, the first warning signal is sent. The first warning signal includes the final accident probability and the warning decision.

[0079] The system obtains the current visibility, current ground friction, current traffic flow, current average vehicle speed, and current average vehicle speed of relevant nodes to obtain the current first accident severity. Based on the current first accident severity, the system obtains the current accident level and the accident level corresponding to key nodes. The system then calculates the difference between the upper limit of the first accident severity corresponding to the accident level and the first accident severity to obtain the third difference value. The system calculates the ratio of the third difference value to the upper limit of the underestimated severity and sets the ratio of the third difference value to the upper limit of the underestimated severity as the accident probability.

[0080] Obtain the current date and the first current time period. If the current date belongs to the time feature or the first current time period, the accident probability is weighted by 0.1. If the current date belongs to the time feature and the first current time period, the accident probability is weighted by 0.2 to obtain the final accident probability.

[0081] The early warning decision includes: setting the expected accident level, obtaining expected parameters based on the first impact model, the expected parameters including expected traffic flow, expected average vehicle speed and the expected average vehicle speed of the corresponding related nodes, and, based on the expected parameters, allowing or limiting the current traffic flow, allowing or reducing the current average vehicle speed, and allowing or reducing the current average vehicle speed of the related nodes, until the expected parameters are met. Specifically, when the current traffic flow of key nodes or related nodes is greater than the expected average speed, traffic is limited; when the current traffic flow is less than or equal to the expected average vehicle speed, no limit is imposed.

[0082] This invention utilizes a two-tiered topology of key nodes and related nodes mined from historical data. This allows for the pre-deployment of AI inference models, cache capacity, and communication bandwidth to potential accident areas, avoiding redundant computation across the entire network. The mapping model from the first accident severity to the accident level is directly embedded in the edge ARM-NPU, enabling local closed-loop decision-making without the need to transmit back to the cloud. Edge nodes periodically transmit new accident samples back to the center for incremental training, and the weights of key nodes are dynamically updated every 24 hours, achieving synchronized resonance between the model and the scenario.

[0083] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A distributed video surveillance integration method based on edge computing, characterized in that, Includes the following steps: Step S100: Based on historical monitoring data, key nodes and related nodes are screened, and the historical monitoring data of key nodes and related nodes are analyzed in the first and second analyses to obtain temporal and spatial characteristics. The key nodes are the three road segments with the highest number of traffic accidents in the first historical time period within the first cycle, and the related nodes are the road segments whose road data should be controlled before the key nodes. The time characteristics include a first change characteristic and a second change characteristic. The first change characteristic of the first quantity of the key node in each first historical time period within the first cycle is obtained through the first analysis, and the second change characteristic of the first quantity of the key node in each first historical time period on the same day is obtained. The historical monitoring data of key nodes and related nodes are subjected to first and second analyses to obtain temporal and spatial characteristics. Select any key node and any historical time period, obtain the first quantity of the key node in the same first historical time period within the first cycle, set the first value as the first quantity threshold, the first value represents the average of the first quantity of the historical time period, compare each first quantity with the first value, when the first quantity is greater than or equal to the first value, retain the corresponding date, when the first quantity is less than the first value, delete the corresponding date, traverse the retained dates corresponding to each historical time period, count the number of times the retained dates appear, sort the number of times the retained dates appear in descending order, set the first number as the number threshold, compare the number of times the retained dates appear with the first number, when the number of times the retained dates appear is greater than or equal to the first number, set the retained dates as accident-prone dates, and set the accident-prone dates as the first change feature; If the number of times a date is retained is less than the number of times it appears the first time, the retained date will be deleted and the process will move on to the next retained date. Step S200: Obtain the first accident severity based on time and space characteristics, and obtain the accident level based on the first accident severity; Step S300: Perform a third analysis on historical monitoring data and accident levels to obtain a first impact model, and send a first early warning signal based on the first impact model.

2. The distributed video surveillance integration method based on edge computing as described in claim 1, characterized in that: The historical monitoring data includes historical time, historical natural data, and historical road data; The historical time is referred to as the historical date; The historical natural data includes visibility and ground friction. The historical road data includes historical road segments, historical traffic flow, historical average vehicle speed, historical minimum vehicle spacing, and corresponding historical license plate numbers. The historical traffic flow is represented by the traffic flow within a first historical time period. The historical average vehicle speed is represented by the average of the average speeds of all vehicles on the road within the first historical time period. The historical minimum vehicle spacing is represented by the minimum distance between the front and rear of any two adjacent vehicles in the same lane on the road within the first historical time period. The historical license plate numbers are the license plate numbers of two or more vehicles belonging to the historical minimum vehicle spacing.

3. The distributed video surveillance integration method based on edge computing as described in claim 2, characterized in that: Before acquiring historical road data, a road segment cutting method is configured, which includes: Obtain the length of the historical road, set the first length as the dividing distance, divide the historical road according to the first length to obtain N historical road segments, and number the historical road segments starting from any end point on either side of the road. The number is a natural number and the number is distributed from 1 to N.

4. The distributed video surveillance integration method based on edge computing as described in claim 1, characterized in that: Methods for filtering key and related nodes based on historical monitoring data include: Select any historical road segment and obtain the historical minimum vehicle spacing corresponding to the historical road segment. Compare the historical minimum vehicle spacing with 0. When the historical minimum vehicle spacing is greater than 0, jump to the next first historical time period of the historical road segment and repeat the first comparison process until the historical minimum vehicle spacing is less than or equal to 0. Obtain the corresponding historical license plate number, count the first number of historical license plate numbers, traverse the first number of the historical road segment in the first historical time period within the first cycle, and calculate the first sum of the first number. The first cycle is represented as one year. Traverse each historical road segment, calculate the first sum of each historical road segment, sort the sums in descending order, select the top three first sums, obtain the historical road segments corresponding to the top three first sums, set the historical road segments corresponding to the top three first sums as key nodes, and set the previous historical road segment of the key node as a related node according to the number order.

5. The distributed video surveillance integration method based on edge computing as described in claim 1, characterized in that: Obtain the first quantity of the key node in each first historical time period on the same day, calculate the first difference between adjacent first quantities in chronological order, traverse each first difference, select the first historical time period corresponding to the first difference with the largest value, traverse the first historical time periods corresponding to each first difference with the largest value in the first period, count the occurrence frequency of each first historical time period, sort the occurrence frequency of each first historical time period in descending order, select the first historical time period corresponding to the occurrence frequency with the largest value, record it as the accident-prone time period, and set the accident-prone time period as the second change feature.

6. The distributed video surveillance integration method based on edge computing as described in claim 1, characterized in that: The second analysis yielded a third variation characteristic between the number of relevant nodes and the number of key nodes in different historical time periods on different days within the first cycle. Get any point in time in the first period, get the first number of any key node and the first number of the corresponding related nodes, calculate the second difference between the first number of key nodes and the first number of the corresponding related nodes, iterate through each second difference in any day, calculate the average of the second differences, iterate through the average of the second differences in each day in the first period, calculate the average of the average of the second differences, record the average of the average of the second differences as the number of related accidents, and set the number of related accidents as the third change feature.

7. The distributed video surveillance integration method based on edge computing as described in claim 1, characterized in that: The accident level is determined based on the severity of the first accident, and the first boundary value is determined based on historical monitoring data. Accident severity levels are categorized into three levels: Level 1, Level 2, and Level 3. The severity of accidents increases progressively from Level 1 to Level 3. Methods for determining the Level 1 accident severity based on temporal and spatial characteristics include: The number of days of the first change characteristic is counted and recorded as the second quantity. The total duration of the second change characteristic is counted and converted into days and recorded as the third quantity. The number of related accidents of the third related quantity is recorded as the fourth quantity. The second sum of the second quantity, the third quantity, and the fourth quantity is calculated and set as the first accident severity. Set the second and third values ​​as the second sum threshold, where the second value is less than the third value, and compare the second sum with the second sum threshold; When the second sum is less than or equal to the second value, the accident level is set to Level 1. When the second sum is greater than the second value and less than or equal to the third value, the accident level is set to Level 2. When the second sum is greater than the third value, the accident level is set to Level 3.

8. The distributed video surveillance integration method based on edge computing as described in claim 1, characterized in that: The methods for constructing the first-influence model include: Obtain the visibility, ground friction, historical traffic flow, historical average vehicle speed, and historical average vehicle speed of any key node, and the corresponding related nodes. Using the severity of the first accident as the dependent variable and visibility, ground friction, historical traffic flow, and historical average vehicle speed as independent variables, perform a multiple regression analysis to obtain the first impact model.

9. The distributed video surveillance integration method based on edge computing as described in claim 8, characterized in that: The methods for sending a first warning signal based on the first impact model include: The system obtains the current road segment that any vehicle is traveling on. If the current road segment is not a key node, no first warning signal is sent. If the current road segment is a key node, the first warning signal is sent. The first warning signal includes the final accident probability and the warning decision. The system obtains the current visibility, current ground friction, current traffic flow, current average vehicle speed, and current average vehicle speed of relevant nodes to obtain the current first accident severity. Based on the current first accident severity, the system obtains the current accident level and the accident level corresponding to key nodes. The system then calculates the difference between the upper limit of the first accident severity corresponding to the accident level and the first accident severity to obtain the third difference value. The system calculates the ratio of the third difference value to the upper limit of the underestimated severity and sets the ratio of the third difference value to the upper limit of the underestimated severity as the accident probability. Obtain the current date and the first current time period. If the current date belongs to the time feature or the first current time period, the accident probability is weighted by 0.

1. If the current date belongs to the time feature and the first current time period, the accident probability is weighted by 0.2 to obtain the final accident probability. The early warning decision includes: setting the expected accident level, obtaining expected parameters based on the first impact model, the expected parameters including expected traffic flow, expected average vehicle speed and the expected average vehicle speed of the corresponding related nodes, and, based on the expected parameters, allowing or limiting the current traffic flow, allowing or reducing the current average vehicle speed, and allowing or reducing the current average vehicle speed of the related nodes, until the expected parameters are met. Specifically, when the current traffic flow of key nodes or related nodes is greater than the expected average speed, traffic is limited; when the current traffic flow is less than or equal to the expected average vehicle speed, no limit is imposed.

Images