Parking lot emergency early warning method and system based on multi-point edge calculation

Abstract

The invention discloses a parking lot emergency early warning method and system based on multi-point edge calculation, and the method comprises the steps: collecting multi-modal environment data and video streams in different regions of a parking lot, obtaining the collected data, carrying out the time synchronization and dynamic threshold filtering of the collected data, and obtaining the preprocessed data; performing spatial-temporal feature alignment and noise filtering on the preprocessed data to obtain a three-dimensional environment situation map; according to the invention, functions of real-time monitoring, analysis, early warning and response of fire, gas leakage and equipment fault emergencies in the parking lot are realized, and through distributed data processing and collaborative analysis, the safety of the parking lot can be significantly improved and an emergency response strategy can be optimized. Meanwhile, the false alarm rate of emergencies can be reduced through multi-modal data fusion and dynamic threshold collaborative analysis, the average response time is shortened, the early warning precision and timeliness are remarkably improved, and the method is suitable for being widely popularized and used.

Description

Technical Field

[0001] The present invention relates to the field of intelligent security technology, and in particular to a parking lot emergency warning method and system based on multi-point edge computing. Background Art

[0002] With the acceleration of urbanization, parking lots, as important public facilities, play an important role in traffic diversion and storage, especially in commercial areas, transportation hubs, residential areas, etc. However, as closed or semi-closed spaces, parking lots often face a series of safety hazards, such as fire, gas leakage, equipment failure and environmental abnormalities.

[0003] The existing parking lot safety monitoring system generally adopts a centralized monitoring architecture, which leads to a high reliance on central servers for data processing and analysis. This centralized architecture faces performance bottlenecks. Since the parking lot safety monitoring system relies on the central server to process global data, the data collected by the sensor must be transmitted to the remote server through the network, which not only leads to a high response delay, but also makes the average response time often exceed 2 seconds when an emergency such as a fire occurs, making it difficult to meet the real-time safety requirements of the parking lot. At the same time, the centralized system also has the risk of single point failure. Once the central server goes down, the entire monitoring system will fail, resulting in the inability to ensure the safety of the parking lot. The data analysis capabilities of the existing parking lot monitoring system are insufficient, especially in static threshold detection. Most of the existing parking lot monitoring systems use fixed thresholds to judge abnormalities, but the parking lot environment is affected by the temperature difference between day and night and vehicle exhaust factors, resulting in a false alarm rate of up to 18%-25%. At present, the existing parking lot safety monitoring systems generally lack the ability to perform spatiotemporal fusion analysis of multi-source heterogeneous data, such as environmental parameters, videos, and equipment status, which makes it impossible to fully grasp the environmental situation of the parking lot. In addition, the emergency response strategy is single and cannot achieve the coordinated optimization of hierarchical warning and dynamic path planning, which easily leads to delays in emergency response. At the same time, the edge computing resource utilization efficiency is low and it cannot support multi-model collaborative reasoning and incremental learning, which seriously limits flexibility and adaptability. Therefore, it is necessary to design a parking lot emergency warning method and system based on multi-point edge computing. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and to better and effectively solve the problems that the existing parking lot safety monitoring system generally lacks the ability to perform spatiotemporal fusion analysis of multi-source heterogeneous data, such as environmental parameters, videos and equipment status, which leads to the inability to fully grasp the environmental situation of the parking lot, and the emergency response strategy is single and cannot achieve the coordinated optimization of hierarchical warning and dynamic path planning, which easily leads to the delay of emergency response. At the same time, the edge computing resource utilization efficiency is low and it cannot support multi-model collaborative reasoning and incremental learning, which seriously limits the flexibility and adaptability. A parking lot emergency warning method and system based on multi-point edge computing is provided, which realizes the functions of real-time monitoring, analysis, warning and response to fire, gas leakage and equipment failure emergencies in the parking lot, and can significantly improve the safety of the parking lot and optimize the emergency response strategy through distributed data processing and collaborative analysis. At the same time, the false alarm rate of emergencies can be reduced through multimodal data fusion and dynamic threshold collaborative analysis, which not only shortens the average response time, but also significantly improves the accuracy and timeliness of the warning.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: A parking lot emergency warning method based on multi-point edge computing includes the following steps: Step A, collecting multimodal environmental data and video streams in different areas of the parking lot and obtaining collected data; Step B, performing time synchronization and dynamic threshold filtering on the collected data and obtaining pre-processed data; Step C, aligning the spatiotemporal features and filtering the noise of the preprocessed data to obtain a three-dimensional environmental situation map; Step D, using multi-model collaborative analysis to determine the type of emergency and warning level based on the three-dimensional environmental situation map; Step E: triggering differentiated emergency response strategies according to the warning level to complete the parking lot emergency warning operation; In step F, knowledge distillation is used to compress the updated model and retain key weights when the model is updated.

[0006] The aforementioned parking lot emergency warning method based on multi-point edge computing, step A, collects multimodal environmental data and video streams in different areas of the parking lot and obtains collected data, wherein the collection of multimodal environmental data and video streams specifically uses a composite temperature and humidity sensor, a gas concentration sensor, a vibration perception sensor and a video collector to collect corresponding data; The composite temperature and humidity sensor uses a PT100 platinum resistor and a capacitive humidity sensor to measure and collect temperature and humidity respectively; The gas concentration sensor uses an electrochemical sensor and an infrared absorption sensor to detect carbon monoxide concentration and hydrogen concentration respectively. The gas concentration sensor uses a dual threshold trigger mechanism combining a carbon monoxide concentration threshold and a hydrogen concentration threshold. The vibration sensing sensor uses a three-axis accelerometer to detect vehicle collisions and abnormal equipment vibrations. The vibration sensing sensor can distinguish between normal vehicle driving and structural vibration events in combination with frequency domain analysis FFT; The video collector adopts a multi-camera collaborative architecture combining a main camera and an auxiliary camera and is provided with a video collection trigger mechanism, wherein the main camera is a wide-angle lens, and the auxiliary camera is an infrared thermal imaging camera. The video collection trigger mechanism specifically switches the video collector to a 60fps frame rate mode when the environmental sensor detects an abnormality.

[0007] In the aforementioned parking lot emergency warning method based on multi-point edge computing, step B performs time synchronization and dynamic threshold filtering on the collected data and obtains pre-processed data, wherein the specific process of time synchronization and dynamic threshold filtering is to use a spatiotemporal alignment module, a noise filtering module and a dynamic threshold calculation module to process the collected data; The time-space alignment module is used to add a unified timestamp to the composite temperature and humidity sensor data, gas concentration sensor data, vibration sensor data and video collector data using the GPS module and NTP protocol, and to compensate for the data timing deviation caused by transmission delay using the interpolation method; The noise filtering module is used to use wavelet transform to remove noise from temperature data and use Kalman filtering to suppress environmental instantaneous fluctuation interference on gas concentration data; The dynamic threshold calculation module is used to calculate the mean μ and standard deviation σ of the environmental parameters based on the sliding window statistics, and then calculate the dynamic adjustment abnormality judgment threshold μ±3σ, while superimposing the day and night cycle correction coefficient.

[0008] The aforementioned parking lot emergency warning method based on multi-point edge computing, step C, performs spatiotemporal feature alignment and noise filtering on the pre-processed data to obtain a three-dimensional environmental situation map. The specific steps are as follows Step C1, add timestamps to the preprocessed data and synchronize them to a unified time axis. Specifically, the IEEE 1588 time protocol is used to achieve ±1ms synchronization accuracy in the local area network, as shown in formula (1): (1) in, is the standard timestamp after time and space calibration, is the original acquisition timestamp, is the time offset calculated by exchanging messages between the master and slave clocks, d is the physical transmission distance, and c is the signal propagation speed; Step C2, using Kalman filtering to eliminate sensor noise, specifically, performing a fifth-order filtering iteration on the preprocessed data, as shown in formula (2): (2) in, and are the state estimation vectors at time k and time k-1 respectively, is the Kalman gain, is the state transfer matrix, is the actual observation value for the kth time, is the observation matrix; Step C3, generating a three-dimensional environmental parameter distribution model based on a spatial interpolation algorithm, wherein the spatial variation function in the spatial interpolation algorithm adopts a spherical model, as shown in formula (3): (3) Where h is the spatial lag distance, is the nugget effect value, C is the sill value, and a is the range; Step C4, superimposing the video stream data and the three-dimensional model to generate a three-dimensional environment situation map, wherein superimposing the video stream data and the three-dimensional model specifically adopts affine transformation to perform coordinate mapping.

[0009] The aforementioned parking lot emergency warning method based on multi-point edge computing, step D, uses multi-model collaborative analysis to determine the emergency type and warning level according to the three-dimensional environmental situation map, specifically using a fire identification model, a gas leakage classification model and an equipment fault detection model to determine the emergency type and warning level through a multi-model voting mechanism, wherein the multi-model voting mechanism specifically sets a confidence threshold and sets weights for the fire identification model, the gas leakage classification model and the equipment fault detection model.

[0010] In the aforementioned parking lot emergency warning method based on multi-point edge computing, the fire recognition model in step D specifically embeds the Backbone end of the YOLOv5 model into the CBAM attention module and sets the channel compression ratio to 16:1; The input features of the gas leakage classification model in step D include gas concentration gradient features, humidity change rate features and vibration spectrum features; The gas concentration gradient characteristic is used to determine the leakage diffusion trend, and the gas concentration gradient characteristic calculation process is shown in formula (4): (4) in, is the gas concentration gradient, is the gas concentration at the current moment, It is the historical concentration value 10 seconds ago; The calculation process of the humidity change rate characteristic is shown in formula (5): (5) in, is the humidity change rate, is the current relative humidity value, is the time interval, is the current humidity value, It is the humidity value 60 seconds ago; The vibration spectrum feature is used to distinguish leakage from conventional vibration by using the ratio of high-frequency energy to low-frequency energy, and the vibration spectrum feature calculation process is shown in formula (6): (6) in, is the vibration energy ratio, P(f) is the power spectral density of the vibration signal, the numerator integration range of 20Hz-50Hz is the high-frequency characteristic frequency band of gas leakage, and the denominator integration range of 5Hz-15Hz is the low-frequency characteristic frequency band of equipment mechanical vibration.

[0011] The aforementioned parking lot emergency warning method based on multi-point edge computing, step E, triggers differentiated emergency response strategies according to the warning level, completes the parking lot emergency warning operation, and specifically includes a local response module, a global evacuation module and a response strategy selection module. The local response module is used to control the start and stop of the sprinkler equipment and the ventilation equipment, the global evacuation module is used to dynamically adjust the LED guide screen and the vehicle navigation information, and the response strategy selection module is used to assign response action priorities according to the warning level and real-time data.

[0012] The aforementioned parking lot emergency warning method based on multi-point edge computing, step F, uses the knowledge distillation method to compress the updated model issued and retain the key weights when the model is updated. The specific steps are as follows: Step F1: compress the updated model sent by the knowledge distillation method. The input of the knowledge distillation method is the updated model package, and the output of the knowledge distillation method is the student model adapted to the hardware resources. , the dynamic distillation loss function of the knowledge distillation method is shown in formula (7), (7) in, is the overall objective function, is the cross entropy loss, y is the true label, Update model package data for input, For the teacher model, is the knowledge distillation loss; Step F2, retain the key weights when updating the model, specifically using the elastic weight screening method to calculate the diagonal elements of the Fisher information matrix of the old model parameters And generate weight importance mask based on median, key weight Regularization constraints are used during optimization, and the key weights As shown in formula (8), (8) in, is the median of data set F; Step F3, construct the triggering conditions and update process of the update model. The specific steps are as follows: Step F31, constructing a trigger condition for updating the model, wherein the trigger condition for updating the model is determined by the change in the accuracy of the validation set, specifically, the update process is initiated when the accuracy of the local validation set of the new model decreases by more than a set probability compared with the current model; Step F32, construct an update process for updating the model. The update process for updating the model adopts a block update method to update in stages according to the priority order of the fully connected layer, the deep convolution layer and the shallow convolution layer, and combines quantization and weight sparsification to compress the model memory usage. The quantization formula is shown in formula (9): (9) in, is the quantized weight, To convert a floating point result to the nearest integer, is the original weight matrix, is the original weight mean, is the original weight standard deviation.

[0013] A parking lot emergency warning system based on multi-point edge computing includes a data acquisition unit, a data preprocessing unit, a multi-model collaborative analysis unit, an emergency response unit and a model update unit, wherein the data acquisition unit is used to collect multimodal environmental data and video streams in different areas of the parking lot and obtain the collected data; the data preprocessing unit is used to perform time synchronization and dynamic threshold filtering on the collected data and obtain the preprocessed data; the data fusion unit is used to perform spatiotemporal feature alignment and noise filtering on the preprocessed data and obtain a three-dimensional environmental situation map; the multi-model collaborative analysis unit is used to determine the type of emergency and the warning level by multi-model collaborative analysis according to the three-dimensional environmental situation map; the emergency response unit is used to trigger differentiated emergency response strategies according to the warning level and complete the parking lot emergency warning operation; the model update unit is used to compress the updated model issued by the knowledge distillation method and retain the key weights when the model is updated.

[0014] The aforementioned parking lot emergency warning system based on multi-point edge computing, the data acquisition unit, the data preprocessing unit and the model update unit are arranged in the distributed edge nodes, the multi-model collaborative analysis unit and the emergency response unit are arranged in the central server, and the distributed edge nodes and the central server are connected by 5G-V2X communication.

[0015] The beneficial effects of the present invention are: (1) A parking lot emergency warning method and system based on multi-point edge computing of the present invention first collects multimodal environmental data and video streams in different areas of the parking lot and obtains collected data, then performs time synchronization and dynamic threshold filtering on the collected data and obtains pre-processed data, then performs spatiotemporal feature alignment and noise filtering on the pre-processed data and obtains a three-dimensional environmental situation map, then uses multi-model collaborative analysis to determine the type of emergency and the warning level based on the three-dimensional environmental situation map, then triggers differentiated emergency response strategies based on the warning level and completes the parking lot emergency warning operation, and finally uses the knowledge distillation method to compress the updated model issued and retain the key weights when the model is updated; the parking lot emergency warning method and system effectively realize the function of using dynamic threshold calculation mechanism and spatiotemporal correlation analysis of environmental parameters to construct an adaptive anomaly detection system, which not only overcomes the contradiction between sensitivity and misjudgment of traditional fixed threshold systems in complex parking lot environments, It also significantly improves the ability to capture hidden safety hazards, and realizes the function of forming a cross-validation emergency judgment logic based on multi-model collaborative analysis and voting decision-making mechanism combined with multi-source information of visual features, environmental parameters and equipment status, which greatly improves the robustness of warning decisions in complex scenarios. It also realizes the function of establishing a hierarchical data processing pipeline and ensuring the integrity of key information under the condition of local preprocessing and feature extraction of edge nodes, which not only enables data to be refined and transmitted, but also builds a low-latency and highly reliable emergency response channel. The model optimization architecture based on knowledge distillation can break through the computing power limitations of edge devices while maintaining analysis accuracy and realize seamless deployment of lightweight AI models on heterogeneous nodes, thereby forming an elastically scalable intelligent perception network, and then supporting plug-and-play and collaborative networking of edge nodes in parking lots of different sizes. This dynamic resource scheduling mechanism can achieve adaptive matching of the scale of parking lot emergency warning system and scenario requirements.

[0016] (2) The present invention can reduce the false alarm rate of emergencies through multimodal data fusion and dynamic threshold collaborative analysis, which not only shortens the average response time, but also significantly improves the accuracy and timeliness of early warning. At the same time, it also realizes the adaptive continuous learning of edge nodes and reduces the maintenance cost of the system.

[0017] (3) The present invention utilizes multiple sensor data and intelligent algorithms for real-time monitoring, analysis and early warning, and combines the edge-cloud collaborative hierarchical early warning architecture to achieve rapid and accurate handling of complex emergencies in parking lots. This multi-node data fusion, dynamic model compression and conflict resolution mechanism greatly improves the safety and management efficiency of parking lots, and reduces the impact of emergencies on property and personnel. The present invention realizes the function of real-time monitoring, analysis, early warning and response to fire, gas leakage and equipment failure emergencies in parking lots, and through distributed data processing and collaborative analysis, it can significantly improve parking lot safety and optimize emergency response strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is an overall flow chart of a parking lot emergency warning method based on multi-point edge computing of the present invention; Figure 2 It is a schematic diagram of the data acquisition principle of the data acquisition unit of the present invention; Figure 3 It is a schematic diagram of the operating principle of the central server of the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further described below in conjunction with the accompanying drawings.

[0020] like Figure 1 As shown, a parking lot emergency warning method based on multi-point edge computing of the present invention includes the following steps: like Figure 2 As shown, step A, collecting multimodal environmental data and video streams in different areas of the parking lot and obtaining collected data, wherein collecting multimodal environmental data and video streams specifically uses a composite temperature and humidity sensor, a gas concentration sensor, a vibration sensing sensor and a video collector to collect corresponding data; The composite temperature and humidity sensor uses a PT100 platinum resistor and a capacitive humidity sensor to measure and collect temperature and humidity respectively; Among them, the PT100 platinum resistance temperature measurement range is -20℃ to 120℃, the capacitive humidity sensor uses the HS1101LF model and is combined with the RC oscillation circuit to measure in the range of 0-100%RH, and the temperature compensation response time of the capacitive humidity sensor is less than 200ms.

[0021] The gas concentration sensor uses an electrochemical sensor and an infrared absorption sensor to detect carbon monoxide concentration and hydrogen concentration respectively. The gas concentration sensor uses a dual threshold trigger mechanism combining a carbon monoxide concentration threshold and a hydrogen concentration threshold. Among them, the electrochemical sensor uses the TGS5042 CO sensor, the detection range of the TGS5042 CO sensor is 0-1000ppm and is set to 30ppm for 10 seconds to trigger the local sound and light alarm, and when the concentration reaches 200ppm, it directly activates the cloud emergency response; the infrared absorption sensor uses the IR7000 sensor, the IR7000 sensor sets 4%LEL as the 20% safety threshold of the lower explosion limit, and if triggered, the exhaust system is immediately started; The vibration sensing sensor uses a three-axis accelerometer to detect vehicle collisions and abnormal equipment vibrations. The vibration sensing sensor can distinguish between normal vehicle driving and structural vibration events in combination with frequency domain analysis FFT; Among them, the vibration sensing sensor uses the ADXL357 three-axis MEMS accelerometer. The range of the ADXL357 three-axis MEMS accelerometer is set to ±16g and the signal-to-noise ratio can reach at least 80dB at a sampling frequency of 100Hz; the frequency domain analysis FFT performs a 1024-point FFT operation and defines 5Hz-15Hz as the vehicle driving characteristic frequency band, and 20Hz-50Hz as the structural abnormality judgment frequency band.

[0022] The video collector adopts a multi-camera collaborative architecture combining a main camera and an auxiliary camera and is provided with a video collection trigger mechanism, wherein the main camera is a wide-angle lens, and the auxiliary camera is an infrared thermal imaging camera. The video collection trigger mechanism specifically switches the video collector to a 60fps frame rate mode when the environmental sensor detects an abnormality.

[0023] Among them, the main camera adopts IMX307 sensor with 120° wide-angle lens. The main camera outputs 1080P@30fps H.264 encoded video stream when the ambient illumination is greater than 10lux; the auxiliary camera adopts FLIR Lepton3.5 component. The auxiliary camera is automatically activated when the ambient illumination is lower than 5lux or the temperature rise rate in the monitoring area exceeds 2℃ / s. The thermal sensitivity of the auxiliary camera is 50mK level; the intelligent trigger mechanism uses abnormal signals from environmental sensors to activate the high frame rate mode. The high frame rate mode specifically automatically adjusts the video resolution to 720P to maintain a 60fps acquisition rate. Motion detection adopts a three-frame differential algorithm and sets a 15% pixel change threshold. The key frame extraction interval is less than 200ms. The video clips of abnormal events are captured 5 seconds before and after, and the volume of a single clip is compressed to less than 8MB through H.265 encoding.

[0024] Step B, performing time synchronization and dynamic threshold filtering on the collected data and obtaining pre-processed data, wherein the specific process of time synchronization and dynamic threshold filtering is to use a time-space alignment module, a noise filtering module and a dynamic threshold calculation module to process the collected data; The time-space alignment module is used to add a unified timestamp to the composite temperature and humidity sensor data, gas concentration sensor data, vibration sensor data and video collector data using the GPS module and NTP protocol, and to compensate for the data timing deviation caused by transmission delay using the interpolation method; The time-space alignment module uses GPS second pulse to trigger timestamp writing and controls the NTP synchronization accuracy within ±10ms. It uses cubic spline interpolation algorithm to compensate for the maximum 500ms transmission delay and sets a data buffer of 120 sampling points. The noise filtering module is used to use wavelet transform to remove noise from temperature data and use Kalman filtering to suppress environmental instantaneous fluctuation interference on gas concentration data; Among them, the noise filtering module implements db4 wavelet five-layer decomposition processing on the temperature data and retains the approximate coefficient a5 for reconstruction. At the same time, the gas concentration data is applied with Kalman filtering with parameters of Q=0.01 and R=0.1, and the initial covariance matrix is ​​set to P0=1; The dynamic threshold calculation module is used to calculate the mean μ and standard deviation σ of the environmental parameters based on the sliding window statistics, and then calculate the dynamic adjustment abnormality judgment threshold μ±3σ, while superimposing the day and night cycle correction coefficient.

[0025] Among them, the dynamic threshold calculation module adopts a 30-minute sliding window of 1800 sampling points and calculates the mean μ and standard deviation σ of environmental parameters in real time based on the Welford algorithm. After superimposing the day and night correction coefficient, κ=1.0 is taken from 06:00 to 18:00 during the day, and κ=1.2 is adjusted from 18:00 to 06:00 at night. For abnormal judgment, 5 consecutive sampling points must exceed the range of μ±3σ, thereby achieving a false alarm suppression rate of 99.7%.

[0026] Step C: perform spatiotemporal feature alignment and noise filtering on the preprocessed data to obtain a three-dimensional environmental situation map. The specific steps are as follows: Step C1, add timestamps to the preprocessed data and synchronize them to a unified time axis. Specifically, the IEEE 1588 time protocol is used to achieve ±1ms synchronization accuracy in the local area network, as shown in formula (1): (1) in, is the standard timestamp after time and space calibration, is the original acquisition timestamp, is the time offset calculated by exchanging messages between the master and slave clocks, d is the physical transmission distance, and c is the signal propagation speed; Specifically, a 0.1s timestamp resolution is added to the composite temperature and humidity sensor data, and a 33ms timestamp interval is used for the video collector data.

[0027] Step C2, using Kalman filtering to eliminate sensor noise, specifically, performing a fifth-order filtering iteration on the preprocessed data, as shown in formula (2): (2) in, and are the state estimation vectors at time k and time k-1 respectively, is the Kalman gain, is the state transfer matrix, is the actual observation value for the kth time, is the observation matrix; Step C3, generating a three-dimensional environmental parameter distribution model based on a spatial interpolation algorithm, wherein the spatial variation function in the spatial interpolation algorithm adopts a spherical model, as shown in formula (3): (3) Where h is the spatial lag distance, is the nugget effect value, C is the sill value, and a is the range; Step C4, superimposing the video stream data and the three-dimensional model to generate a three-dimensional environment situation map, wherein superimposing the video stream data and the three-dimensional model specifically uses affine transformation to perform coordinate mapping; Among them, the affine transformation matrix parameters are .

[0028] like Figure 3 As shown, step D, according to the three-dimensional environmental situation map, multi-model collaborative analysis is used to determine the type of emergency and the warning level, specifically, the fire recognition model, the gas leakage classification model and the equipment fault detection model are used to determine the type of emergency and the warning level through a multi-model voting mechanism, wherein the multi-model voting mechanism specifically sets a confidence threshold and sets weights for the fire recognition model, the gas leakage classification model and the equipment fault detection model; the fire recognition model specifically embeds the Backbone end of the YOLOv5 model into the CBAM attention module and sets the channel compression ratio to 16:1; the input features of the gas leakage classification model include gas concentration gradient features, humidity change rate features and vibration spectrum features; The multi-model voting mechanism sets the confidence threshold as follows: when the confidence of the output of the two models exceeds 75%, the warning level is upgraded by one level; the weight distribution scheme is that the fire identification model accounts for 45%, the gas leakage classification model accounts for 35%, and the equipment fault detection model accounts for 20%; The gas concentration gradient characteristic is used to determine the leakage diffusion trend, and the gas concentration gradient characteristic calculation process is shown in formula (4): (4) in, is the gas concentration gradient, is the gas concentration at the current moment, It is the historical concentration value 10 seconds ago; The calculation process of the humidity change rate characteristic is shown in formula (5): (5) in, is the humidity change rate, is the current relative humidity value, is the time interval, is the current humidity value, It is the humidity value 60 seconds ago; The vibration spectrum feature is used to distinguish leakage from conventional vibration by using the ratio of high-frequency energy to low-frequency energy, and the vibration spectrum feature calculation process is shown in formula (6): (6) in, is the vibration energy ratio, P(f) is the power spectrum density of the vibration signal, the numerator integration range of 20Hz-50Hz is the high-frequency characteristic frequency band of gas leakage, and the denominator integration range of 5Hz-15Hz is the low-frequency characteristic frequency band of mechanical vibration of the equipment; When the threshold ER is set to ≥ 3.0, it is judged as a leak.

[0029] Step E, triggering differentiated emergency response strategies according to the warning level to complete the parking lot emergency warning operation, specifically including a local response module, a global evacuation module and a response strategy selection module, the local response module is used to control the start and stop of the sprinkler equipment and the ventilation equipment, the global evacuation module is used to dynamically adjust the LED guide screen and the vehicle navigation information, and the response strategy selection module is used to assign the response action priority according to the warning level and real-time data; The local response module control parameter settings include a 3s delay in starting the sprinkler system after it is triggered, water pressure maintained at 0.35±0.05MPa, ventilation equipment wind speed adjustment in three levels, one level increase for every 50ppm increase in CO concentration, and a maximum wind speed of 12m / s. The specific priority order is: fire incident > gas leakage > equipment failure.

[0030] Step F: Use knowledge distillation to compress the updated model and retain key weights when updating the model. The specific steps are as follows: Step F1: compress the updated model sent by the knowledge distillation method. The input of the knowledge distillation method is the updated model package, and the output of the knowledge distillation method is the student model adapted to the hardware resources. , the dynamic distillation loss function of the knowledge distillation method is shown in formula (7), (7) in, is the overall objective function, is the cross entropy loss, y is the true label, Update model package data for input, For the teacher model, is the knowledge distillation loss; Step F2, retain the key weights when updating the model, specifically using the elastic weight screening method to calculate the diagonal elements of the Fisher information matrix of the old model parameters And generate weight importance mask based on median, key weight Regularization constraints are used during optimization, and the key weights As shown in formula (8), (8) in, is the median of data set F; Step F3, construct the triggering conditions and update process of the update model. The specific steps are as follows: Step F31, constructing a trigger condition for updating the model, wherein the trigger condition for updating the model is determined by the change in the accuracy of the validation set, specifically, the update process is initiated when the accuracy of the local validation set of the new model decreases by more than a set probability compared with the current model; Step F32, construct an update process for updating the model. The update process for updating the model adopts a block update method to update in stages according to the priority order of the fully connected layer, the deep convolution layer and the shallow convolution layer, and combines quantization and weight sparsification to compress the model memory usage. The quantization formula is shown in formula (9): (9) in, is the quantized weight, To convert a floating point result to the nearest integer, is the original weight matrix, is the original weight mean, is the original weight standard deviation.

[0031] A parking lot emergency warning system based on multi-point edge computing includes a data acquisition unit, a data preprocessing unit, a multi-model collaborative analysis unit, an emergency response unit and a model update unit, wherein the data acquisition unit is used to collect multimodal environmental data and video streams in different areas of the parking lot and obtain the collected data; the data preprocessing unit is used to perform time synchronization and dynamic threshold filtering on the collected data and obtain the preprocessed data; the data fusion unit is used to perform spatiotemporal feature alignment and noise filtering on the preprocessed data and obtain a three-dimensional environmental situation map; the multi-model collaborative analysis unit is used to determine the type of emergency and the warning level by multi-model collaborative analysis according to the three-dimensional environmental situation map; the emergency response unit is used to trigger differentiated emergency response strategies according to the warning level and complete the parking lot emergency warning operation; the model update unit is used to compress the updated model issued by the knowledge distillation method and retain the key weights when the model is updated.

[0032] Specifically, the data acquisition unit, the data preprocessing unit and the model updating unit are arranged in the distributed edge node, the multi-model collaborative analysis unit and the emergency response unit are arranged in the central server, and the distributed edge node and the central server are connected by 5G-V2X communication.

[0033] In summary, the parking lot emergency warning method and system based on multi-point edge computing of the present invention first collects multimodal environmental data and video streams in different areas of the parking lot and obtains the collected data, then performs time synchronization and dynamic threshold filtering on the collected data and obtains the preprocessed data, then performs spatiotemporal feature alignment and noise filtering on the preprocessed data and obtains a three-dimensional environmental situation map, and then uses multi-model collaborative analysis to determine the type of emergency and the warning level based on the three-dimensional environmental situation map, then triggers differentiated emergency response strategies based on the warning level and completes the parking lot emergency warning operation, and finally uses the knowledge distillation method to compress the updated model issued and retain the key weights when the model is updated; effectively realize The parking lot emergency warning method and system has the function of constructing an adaptive anomaly detection system by using a dynamic threshold calculation mechanism and spatiotemporal correlation analysis of environmental parameters. It not only overcomes the contradiction between sensitivity and misjudgment of traditional fixed threshold systems in complex parking lot environments, but also significantly improves the ability to capture hidden safety hazards. It also realizes the function of forming a cross-validation emergency judgment logic based on multi-model collaborative analysis and voting decision-making mechanism combined with multi-source information of visual features, environmental parameters and equipment status, which greatly improves the robustness of warning decisions in complex scenarios. It also realizes the function of establishing a hierarchical data processing pipeline and ensuring the integrity of key information under the conditions of local preprocessing and feature extraction of edge nodes. It not only allows data to be refined and transmitted, but also builds a low-latency and highly reliable emergency response channel. The model optimization architecture based on knowledge distillation can break through the computing power limitations of edge devices while maintaining analysis accuracy and realize seamless deployment of lightweight AI models on heterogeneous nodes, thereby forming an elastically scalable intelligent perception network, and then supporting plug-and-play and collaborative networking of edge nodes in parking lots of different sizes. This dynamic resource scheduling mechanism can achieve adaptive matching of the scale of parking lot emergency warning system and scenario requirements; the present invention can reduce the false alarm rate of emergencies through multimodal data fusion and dynamic threshold collaborative analysis, which not only shortens the average response time, but also significantly improves the warning accuracy and timeliness, while also realizing edge The adaptive continuous learning of nodes reduces the maintenance cost of the system; the present invention utilizes a variety of sensor data and intelligent algorithms for real-time monitoring, analysis and early warning, and combines the edge-cloud collaborative hierarchical early warning architecture to achieve rapid and accurate handling of complex emergencies in parking lots. This multi-node data fusion, dynamic model compression and conflict resolution mechanism greatly improves the safety and management efficiency of parking lots, and reduces the impact of emergencies on property and personnel. The present invention realizes the function of real-time monitoring, analysis, early warning and response to fire, gas leakage and equipment failure emergencies in parking lots, and through distributed data processing and collaborative analysis, it can significantly improve parking lot safety and optimize emergency response strategies.

[0034] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A parking lot emergency warning method based on multi-point edge computing, characterized by: The following steps are included: Step A, collecting multimodal environmental data and video streams in different areas of the parking lot and obtaining collected data; Step B, performing time synchronization and dynamic threshold filtering on the collected data and obtaining pre-processed data; Step C, aligning the spatiotemporal features and filtering the noise of the preprocessed data to obtain a three-dimensional environmental situation map; Step D, using multi-model collaborative analysis to determine the type of emergency and warning level based on the three-dimensional environmental situation map; Step E: triggering differentiated emergency response strategies according to the warning level to complete the parking lot emergency warning operation; In step F, knowledge distillation is used to compress the updated model and retain key weights when the model is updated.

2. According to claim 1, a parking lot emergency warning method based on multi-point edge computing is characterized in that: Step A, collecting multimodal environmental data and video streams in different areas of the parking lot and obtaining collected data, wherein the collecting of multimodal environmental data and video streams specifically uses a composite temperature and humidity sensor, a gas concentration sensor, a vibration sensing sensor and a video collector to collect corresponding data; The composite temperature and humidity sensor uses a PT100 platinum resistor and a capacitive humidity sensor to measure and collect temperature and humidity respectively; The gas concentration sensor uses an electrochemical sensor and an infrared absorption sensor to detect carbon monoxide concentration and hydrogen concentration respectively. The gas concentration sensor uses a dual threshold trigger mechanism combining a carbon monoxide concentration threshold and a hydrogen concentration threshold. The vibration sensing sensor uses a three-axis accelerometer to detect vehicle collisions and abnormal equipment vibrations. The vibration sensing sensor can distinguish between normal vehicle driving and structural vibration events in combination with frequency domain analysis FFT; The video collector adopts a multi-camera collaborative architecture combining a main camera and an auxiliary camera and is provided with a video collection trigger mechanism, wherein the main camera is a wide-angle lens, and the auxiliary camera is an infrared thermal imaging camera. The video collection trigger mechanism specifically switches the video collector to a 60fps frame rate mode when the environmental sensor detects an abnormality.

3. According to claim 2, a parking lot emergency warning method based on multi-point edge computing is characterized in that: Step B, performing time synchronization and dynamic threshold filtering on the collected data and obtaining pre-processed data, wherein the specific process of time synchronization and dynamic threshold filtering is to use a time-space alignment module, a noise filtering module and a dynamic threshold calculation module to process the collected data; The time-space alignment module is used to add a unified timestamp to the composite temperature and humidity sensor data, gas concentration sensor data, vibration sensor data and video collector data using the GPS module and NTP protocol, and to compensate for the data timing deviation caused by transmission delay using the interpolation method; The noise filtering module is used to use wavelet transform to remove noise from temperature data and use Kalman filtering to suppress environmental instantaneous fluctuation interference on gas concentration data; The dynamic threshold calculation module is used to calculate the mean μ and standard deviation σ of the environmental parameters based on the sliding window statistics, and then calculate the dynamic adjustment abnormality judgment threshold μ±3σ, while superimposing the day and night cycle correction coefficient.

4. According to claim 3, a parking lot emergency warning method based on multi-point edge computing is characterized in that: Step C: perform spatiotemporal feature alignment and noise filtering on the preprocessed data to obtain a three-dimensional environmental situation map. The specific steps are as follows: Step C1, add timestamps to the preprocessed data and synchronize them to a unified time axis. Specifically, the IEEE 1588 time protocol is used to achieve ±1ms synchronization accuracy in the local area network, as shown in formula (1): （1） in, is the standard timestamp after time and space calibration, is the original acquisition timestamp, is the time offset calculated by exchanging messages between the master and slave clocks, d is the physical transmission distance, and c is the signal propagation speed; Step C2, using Kalman filtering to eliminate sensor noise, specifically, performing a fifth-order filtering iteration on the preprocessed data, as shown in formula (2): （2） in, and are the state estimation vectors at time k and time k-1 respectively, is the Kalman gain, is the state transfer matrix, is the actual observation value for the kth time, is the observation matrix; Step C3, generating a three-dimensional environmental parameter distribution model based on a spatial interpolation algorithm, wherein the spatial variation function in the spatial interpolation algorithm adopts a spherical model, as shown in formula (3): （3） Where h is the spatial lag distance, is the nugget effect value, C is the sill value, and a is the range; Step C4, superimposing the video stream data and the three-dimensional model to generate a three-dimensional environment situation map, wherein superimposing the video stream data and the three-dimensional model specifically adopts affine transformation to perform coordinate mapping.

5. A parking lot emergency warning method based on multi-point edge computing according to claim 4, characterized in that: Step D, based on the three-dimensional environmental situation map, a multi-model collaborative analysis is used to determine the type of emergency and the warning level. Specifically, the fire identification model, the gas leakage classification model and the equipment fault detection model are used to determine the type of emergency and the warning level through a multi-model voting mechanism, wherein the multi-model voting mechanism specifically sets a confidence threshold and sets weights for the fire identification model, the gas leakage classification model and the equipment fault detection model.

6. A parking lot emergency warning method based on multi-point edge computing according to claim 5, characterized in that: The fire recognition model in step D specifically embeds the Backbone end of the YOLOv5 model into the CBAM attention module and sets the channel compression ratio to 16:1; The input features of the gas leakage classification model in step D include gas concentration gradient features, humidity change rate features and vibration spectrum features; The gas concentration gradient characteristic is used to determine the leakage diffusion trend, and the gas concentration gradient characteristic calculation process is shown in formula (4): （4） in, is the gas concentration gradient, is the gas concentration at the current moment, It is the historical concentration value 10 seconds ago; The calculation process of the humidity change rate characteristic is shown in formula (5): （5） in, is the humidity change rate, is the current relative humidity value, is the time interval, is the current humidity value, It is the humidity value 60 seconds ago; The vibration spectrum feature is used to distinguish leakage from normal vibration by using the ratio of high-frequency energy to low-frequency energy, and the vibration spectrum feature calculation process is shown in formula (6): （6） in, is the vibration energy ratio, P(f) is the power spectral density of the vibration signal, the numerator integration range of 20Hz-50Hz is the high-frequency characteristic frequency band of gas leakage, and the denominator integration range of 5Hz-15Hz is the low-frequency characteristic frequency band of equipment mechanical vibration.

7. A parking lot emergency warning method based on multi-point edge computing according to claim 6, characterized in that: Step E, triggering differentiated emergency response strategies according to the warning level, completing the parking lot emergency warning operation, specifically includes a local response module, a global evacuation module and a response strategy selection module. The local response module is used to control the start and stop of the sprinkler equipment and ventilation equipment, the global evacuation module is used to dynamically adjust the LED guide screen and vehicle navigation information, and the response strategy selection module is used to assign response action priorities according to the warning level and real-time data.

8. The parking lot emergency warning method based on multi-point edge computing according to claim 7 is characterized by: Step F: Use knowledge distillation to compress the updated model and retain key weights when updating the model. The specific steps are as follows: Step F1: compress the updated model sent by the knowledge distillation method. The input of the knowledge distillation method is the updated model package, and the output of the knowledge distillation method is the student model adapted to the hardware resources. , the dynamic distillation loss function of the knowledge distillation method is shown in formula (7), （7） in, is the overall objective function, is the cross entropy loss, y is the true label, Update model package data for input, For the teacher model, is the knowledge distillation loss; Step F2, retain the key weights when updating the model, specifically using the elastic weight screening method to calculate the diagonal elements of the Fisher information matrix of the old model parameters And generate weight importance mask based on median, key weight Regularization constraints are used during optimization, and the key weights As shown in formula (8), （8） in, is the median of data set F; Step F3, construct the triggering conditions and update process of the update model. The specific steps are as follows: Step F31, constructing a trigger condition for updating the model, wherein the trigger condition for updating the model is determined by the change in the accuracy of the validation set, specifically, the update process is initiated when the accuracy of the local validation set of the new model decreases by more than a set probability compared with the current model; Step F32, construct an update process for updating the model. The update process for updating the model adopts a block update method to update in stages according to the priority order of the fully connected layer, the deep convolution layer and the shallow convolution layer, and combines quantization and weight sparsification to compress the model memory usage. The quantization formula is shown in formula (9): （9） in, is the quantized weight, To convert a floating point result to the nearest integer, is the original weight matrix, is the original weight mean, is the original weight standard deviation.

9. A parking lot emergency warning system based on multi-point edge computing, wherein the specific warning process of the parking lot emergency warning system is based on the method described in any one of claims 1 to 9, characterized in that: It includes a data acquisition unit, a data preprocessing unit, a multi-model collaborative analysis unit, an emergency response unit and a model update unit, wherein the data acquisition unit is used to collect multi-modal environmental data and video streams in different areas of the parking lot and obtain the collected data; The data preprocessing unit is used to perform time synchronization and dynamic threshold filtering on the collected data and obtain preprocessed data; The data fusion unit is used to perform spatiotemporal feature alignment and noise filtering on the pre-processed data and obtain a three-dimensional environmental situation map; The multi-model collaborative analysis unit is used to determine the type of emergency and the warning level by using multi-model collaborative analysis according to the three-dimensional environmental situation map; The emergency response unit is used to trigger differentiated emergency response strategies according to the warning level and complete the parking lot emergency warning operation; The model updating unit is used to compress the updated model sent down by using the knowledge distillation method and retain key weights when the model is updated.

10. The parking lot emergency warning system based on multi-point edge computing according to claim 9, characterized in that: The data acquisition unit, data preprocessing unit and model updating unit are arranged in distributed edge nodes, the multi-model collaborative analysis unit and emergency response unit are arranged in a central server, and a 5G-V2X communication connection is adopted between the distributed edge nodes and the central server.

Images