Method for monitoring abnormal events based on radar and infrared sensor data

Abstract

The embodiment of the application relates to a kind of based on radar and infrared sensing data carries out abnormal event monitoring method, comprising: in the supervision space, place M millimeter wave radar and U infrared sensor, set up three-dimensional grid network and target coordinate system, for radar and infrared sensor respectively set coordinate system mapping rule, temperature measurement coordinate and temperature difference threshold value.Millimeter wave radar generates single frame point cloud periodically scanning, infrared sensor generates temperature measurement data and saves periodically monitoring.Pressure time frequency obtains point cloud sequence and temperature measurement sequence, set subdivision time axis and make data sampling time and it align.According to point cloud sequence, predict target trajectory, according to temperature measurement sequence, predict temperature measurement trajectory, peak temperature and ambient temperature difference.Finally, based on predicted trajectory, personnel stay time abnormal event identification early warning is carried out, according to temperature measurement trajectory, single-point high temperature abnormal event identification early warning is carried out, using period peak temperature and period ambient temperature difference, environmental high temperature abnormal event identification early warning is carried out.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for monitoring abnormal events based on radar and infrared sensor data. Background Technology

[0002] In various regulated spaces, such as industrial plants, large warehouses, and public buildings, security and anomaly monitoring are crucial. Traditional manual inspection methods are not only costly in terms of manpower and resources, but also lack real-time and comprehensive monitoring capabilities, failing to detect potential anomalies in a timely manner and easily leading to safety accidents or increased property damage. Therefore, utilizing advanced technologies to achieve automated and intelligent anomaly monitoring has become an inevitable trend.

[0003] Currently, common anomaly detection technologies are primarily based on single-type sensor data. For example, video surveillance-based monitoring methods capture video images through cameras and use image recognition technology to detect and analyze people, objects, etc., in the footage to identify abnormal behavior or events. However, this method has many limitations in practical applications. Video surveillance is easily affected by lighting conditions; image quality deteriorates significantly at night or in low-light environments, leading to lower recognition accuracy. Simultaneously, the massive volume of video data places high demands on storage and transmission bandwidth, increasing system cost and complexity. Furthermore, video surveillance raises privacy concerns; large-scale deployment of cameras in privacy-sensitive locations may raise public concerns.

[0004] Another common monitoring method is temperature monitoring based on infrared sensors. Infrared sensors can obtain surface temperature information by detecting the infrared radiation emitted by objects, thereby enabling the monitoring of temperature anomalies. However, relying solely on infrared sensors for monitoring also has its limitations. Infrared sensors can only acquire temperature information for localized areas, making it difficult to comprehensively grasp the dynamic position and movement trajectory of targets throughout the entire monitored space. Moreover, the temperature measurement accuracy of infrared sensors is easily affected by environmental factors, such as ambient temperature, humidity, and air movement, leading to deviations in the measurement results and affecting the accuracy of anomaly detection.

[0005] Millimeter-wave radar, as a sensor capable of transmitting and receiving millimeter-wave signals, possesses advantages such as strong penetration, strong anti-interference capability, and immunity to light and smoke, enabling precise target detection and positioning in various complex environments. It can acquire information such as the target's three-dimensional coordinates, velocity, and angle, and by analyzing this data, it can predict and track the target's trajectory. However, millimeter-wave radar also has some limitations when used alone. For example, its target classification ability is relatively weak, making it difficult to accurately distinguish between different types of targets; simultaneously, millimeter-wave radar cannot directly acquire the target's temperature information, making it unable to effectively monitor some temperature-related anomalies, such as fires or equipment overheating.

[0006] In summary, single-sensor monitoring technologies, due to their inherent limitations, are insufficient to meet the demands for comprehensive and accurate monitoring of abnormal events within complex regulatory spaces. Therefore, how to comprehensively utilize the advantages of multiple sensors to achieve the fusion and analysis of multi-source data, thereby improving the accuracy and reliability of abnormal event monitoring, has become a pressing technical problem. This invention proposes a method for abnormal event monitoring based on radar and infrared sensor data, aiming to address the problems existing in the prior art. By fusing data from millimeter-wave radar and infrared sensors, it achieves effective monitoring and early warning of various abnormal events within the regulatory space. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for monitoring abnormal events based on radar and infrared sensor data, thereby solving the problems existing in the prior art.

[0008] To achieve the above objectives, the present invention provides a method for anomaly event monitoring based on radar and infrared sensor data, the method comprising:

[0009] M millimeter-wave radars and U infrared sensors are deployed within the monitored space; a corresponding three-dimensional mesh network is set up for the monitored space; the three-dimensional coordinate system of the three-dimensional mesh network is used as the target coordinate system; based on the radar parameters of each millimeter-wave radar, a coordinate system mapping rule from the radar coordinate system to the target coordinate system is set; based on the installation position of each infrared sensor, a corresponding mesh coordinate is set for each infrared sensor in the target coordinate system as the temperature measurement coordinate; and a corresponding sensor temperature difference threshold is set for each infrared sensor.

[0010] Each millimeter-wave radar periodically scans the monitored space according to its own scanning frequency, generating and saving corresponding single-frame point clouds; each infrared sensor periodically monitors the temperature at its respective installation location according to its own temperature measurement frequency, generating and saving corresponding temperature measurement data.

[0011] At a preset first time frequency, M point cloud sequences generated by M millimeter-wave radars and U temperature measurement sequences generated by U infrared sensors are periodically acquired within the most recent time period T; a corresponding subdivided time axis is set for the current time period T; and multiple subdivided time points are set on the subdivided time axis based on preset time intervals; and it is ensured that each sampling time of all point cloud sequences and each temperature measurement time of all temperature measurement sequences can be aligned with a subdivided time point on the subdivided time axis.

[0012] Multiple first predicted trajectories are obtained by predicting the target trajectory based on M point cloud sequences. Each first predicted trajectory corresponds to a real target in the actual monitoring space. The first predicted trajectory is the activity trajectory of the real target in the time period and monitoring space. The first predicted trajectory is formed by sequentially sorting multiple first predicted trajectory points. Each first predicted trajectory point includes a first time point, first center point coordinates, first three-dimensional size, first orientation angle, and first target type. The first time point corresponds one-to-one with the subdivided time points. The first center point coordinates are three-dimensional grid coordinates in a three-dimensional grid network.

[0013] Based on U temperature measurement sequences, the corresponding U first temperature measurement trajectories, peak temperatures, and ambient temperature differences are predicted to obtain a set of peak temperatures and ambient temperature differences for each time period. Among them, the first temperature measurement trajectory includes multiple first temperature measurement trajectory points. Each first temperature measurement trajectory point includes a first temperature measurement time point, a first temperature measurement coordinate, and a first temperature. The total number of trajectory points of each first temperature measurement trajectory is consistent with the total number of subdivided time points of the subdivided time axis.

[0014] Based on all the first predicted trajectories obtained this time, we will identify and issue warnings for abnormal events related to the time of personnel stay.

[0015] Based on the first temperature measurement trajectories obtained this time, single-point high temperature anomaly events are identified and early warnings are issued;

[0016] Based on the peak temperature and ambient temperature difference obtained during this period, we will identify and issue early warnings for abnormal high-temperature events.

[0017] The method for anomaly monitoring based on radar and infrared sensor data provided by this invention has the following technical effects:

[0018] Multi-source data fusion enhances monitoring accuracy and reliability. By combining point cloud data from millimeter-wave radar (providing the target's three-dimensional position, size, orientation, and trajectory) with temperature data from infrared sensors (providing temperature distribution and dynamic changes), multi-dimensional perception of targets within the monitored space is achieved. Compared to single-sensor solutions, this method effectively avoids monitoring failures caused by misjudgments or obstructions from a single data source, significantly improving the accuracy and robustness of anomaly identification.

[0019] A spatiotemporal alignment mechanism ensures collaborative data analysis. By establishing a three-dimensional grid network as a unified target coordinate system and designing mapping rules from the radar coordinate system to the target coordinate system, spatial alignment of data from multiple radars and infrared sensors is achieved. Simultaneously, by subdividing the time axis and using time point alignment strategies, the synchronization of point cloud sequences and temperature measurement sequences in the time dimension is ensured. This mechanism provides a precise spatiotemporal reference for subsequent trajectory prediction and anomaly event correlation analysis, avoiding false alarms or missed detections caused by data misalignment.

[0020] Dynamic trajectory prediction enables refined anomaly identification. Target trajectory prediction based on point cloud sequences can generate a first predicted trajectory containing parameters such as time point, grid coordinates, and 3D dimensions, accurately depicting the real-time activity status of targets within the monitored space. Combining the peak temperature of the temperature measurement trajectory with the analysis of the environmental temperature difference, this method can distinguish different types of abnormal events, such as personnel lingering (based on trajectory continuity), single-point high temperature (based on local temperature abrupt changes), and environmental high temperature (based on global temperature difference thresholds), achieving refined identification from "existence of an anomaly" to "anomaly type."

[0021] A tiered early warning mechanism improves emergency response efficiency. This method sets up independent identification and early warning logic for three types of abnormal events: personnel confinement, single-point high temperatures, and ambient high temperatures, forming a tiered response system. For example, abnormal personnel confinement time can be triggered by the duration of trajectory points; single-point high temperature anomalies can be judged by combining sensor temperature difference thresholds and temperature rise rates; and ambient high temperature anomalies are triggered by a linkage between peak temperature over a time period and dynamic ambient temperature difference thresholds. This tiered strategy prioritizes high-risk events and optimizes resource allocation efficiency.

[0022] Adaptive parameter configuration enhances scenario adaptability. By setting adjustable parameters such as the sensor temperature difference threshold for the infrared sensor and configuring dynamic time intervals for subdivided time axes, this method can adapt to the monitoring needs of different regulatory scenarios (such as indoor / outdoor, dense / sparse spaces). For example, in densely populated scenarios, the time interval can be shortened to increase the trajectory prediction frequency, and in high-temperature industrial environments, the low temperature difference threshold can be adjusted to improve sensitivity to environmental anomalies, thereby balancing monitoring accuracy and computational resource consumption.

[0023] Low-latency real-time monitoring meets safety control requirements. Each sensor independently collects data at its inherent frequency, and the data is periodically aggregated and analyzed at a preset time frequency, ensuring data integrity while reducing the real-time processing pressure on the system. A detailed time axis and alignment mechanism ensure that the latency of abnormal event identification is controllable (depending only on the time frequency setting), meeting the stringent real-time monitoring requirements of scenarios such as public safety and industrial production. Attached Figure Description

[0024] Figure 1This is a schematic diagram of the method for abnormal event monitoring based on radar and infrared sensor data provided in an embodiment of the present invention;

[0025] Figure 2 A schematic diagram showing the deployment locations of radar and infrared sensors;

[0026] Figure 3 This is a schematic diagram of a 3D mesh network and a 3D mesh.

[0027] Figure 4 for Figure 1 The detailed flowchart of step 140;

[0028] Figure 5 To obtain a schematic diagram of the target trajectory set;

[0029] Figure 6 To obtain another schematic diagram of the target trajectory set;

[0030] Figure 7 for Figure 1 The detailed flowchart of step 150;

[0031] Figure 8 The processing flowchart for the first temperature lattice is shown below.

[0032] Figure 9 for Figure 1 The detailed flowchart of step 160;

[0033] Figure 10 for Figure 1 The detailed flowchart of step 170. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0035] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0036] Figure 1 This is a schematic flowchart illustrating a method for abnormal event monitoring based on radar and infrared sensor data, provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0037] Step 110: Deploy M millimeter-wave radars and U infrared sensors within the monitored space; set up a corresponding three-dimensional mesh network for the monitored space; use the three-dimensional coordinate system of the three-dimensional mesh network as the target coordinate system; based on the radar parameters of each millimeter-wave radar, set a coordinate system mapping rule from the radar coordinate system to the target coordinate system; and based on the installation position of each infrared sensor, set a corresponding mesh coordinate in the target coordinate system as the temperature measurement coordinate for each infrared sensor; and set a corresponding sensor temperature difference threshold for each infrared sensor.

[0038] Specifically, this application relates to a method for monitoring abnormal events based on radar and infrared sensor data. The radar can be a millimeter-wave radar, and the sensor deployment method is described in [reference needed]. Figure 2 The deployment rules for millimeter-wave radar are as follows: it is deployed at the side / vertices of the monitored space; the deployment rules for infrared sensors are as follows: the monitored space is divided into a three-dimensional grid, and sensors are deployed at the grid locations where they are needed. The coordinates of the center point of the grid where the sensor is located are used as the sensor coordinates.

[0039] See 3D mesh network Figure 3 The regulatory space is divided into a three-dimensional grid, resulting in a three-dimensional grid network V; V consists of multiple three-dimensional grids v. x,y,z The structure consists of (x, y, z) representing the grid coordinates of the 3D mesh v. The 3D coordinate system of the 3D mesh network V is used as the target coordinate system.

[0040] For each millimeter-wave radar, a coordinate system mapping rule f is established between its corresponding radar coordinate system and the target coordinate system. R→V (In reality, it's a set of coordinate transformation matrices); based on the coordinate system mapping rules f of each radar. R→V The position coordinates (x, y) of each radar point cloud can be obtained. * ,y * ,z * Convert f to grid coordinates (x, y, z) in the target coordinate system: R→V (x * ,y * ,z * ) = (x, y, z).

[0041] For each infrared sensor, the grid coordinates corresponding to its installation location can be used as the sensor coordinates / temperature measurement coordinates.

[0042] Step 120: Each millimeter-wave radar periodically scans the monitored space according to its own scanning frequency to generate and save the corresponding single-frame point cloud; each infrared sensor periodically monitors the temperature at its respective installation location according to its own temperature measurement frequency to generate and save the corresponding temperature measurement data.

[0043] Each frame of point cloud fed back by millimeter-wave radar contains multiple spatial point data, and the data fields of each point data are shown in Table 1:

[0044] Table 1

[0045]

[0046] The infrared sensor is a non-imaging type. The temperature data fields returned by the infrared sensor are shown in Table 2:

[0047] Table 2

[0048] timestamp HH:MM:SS or others temperature Celsius

[0049] By transforming and setting coordinates, new point cloud data fields and temperature data fields can be obtained, as shown in Tables 3 and 4 respectively:

[0050] Table 3

[0051]

[0052]

[0053] Table 4

[0054] timestamp HH:MM:SS or others Coordinates (x, y, z) 3D coordinates of the target coordinate system temperature Unit: degrees Celsius

[0055] Step 130: Periodically acquire M point cloud sequences generated by M millimeter-wave radars and U temperature measurement sequences generated by U infrared sensors within the most recent time period T according to a preset first time frequency; set a corresponding subdivided time axis for the current time period T; set multiple subdivided time points on the subdivided time axis based on preset time intervals; and ensure that each sampling time of all point cloud sequences and each temperature measurement time of all temperature measurement sequences can be aligned with a subdivided time point on the subdivided time axis.

[0056] Specifically, M point cloud sequences generated by M millimeter-wave radars within the most recent time period T are periodically processed according to a preset first time frequency. and U temperature measurement sequences generated by U infrared sensors The acquisition is performed; a corresponding subdivision time axis L is set for the current time period T; multiple subdivision time points t are set on the subdivision time axis L based on a preset time interval Δt; and all point cloud sequences are ensured to be acquired. Each sampling time and all temperature measurement sequences Each temperature measurement time can be aligned with a subdivided time point t on the subdivided time axis L.

[0057] Step 140: Based on the M point cloud sequences, predict the target trajectory to obtain multiple corresponding first predicted trajectories.

[0058] Among them, based on M point cloud sequences Multiple first predicted trajectories are obtained by predicting the target trajectory. Each first predicted trajectory corresponds to a real target within the actual regulatory space; the first predicted trajectory is the activity trajectory of the real target within the time period and regulatory space; the first predicted trajectory is formed by sequentially sorting multiple first predicted trajectory points; each first predicted trajectory point includes a first time point, first center point coordinates, first three-dimensional dimensions, first orientation angle, and first target type; the first time point corresponds one-to-one with the subdivided time points; the first center point coordinates are three-dimensional grid coordinates in a three-dimensional grid network.

[0059] like Figure 4 As shown, step 140 includes the following:

[0060] Step 1401: Obtain M point cloud sequences from M millimeter-wave radars over time period T. Where 1 ≤ radar index i ≤ M; each point cloud sequence D i Including multiple first-frame point clouds 1≤time index j≤N i N i For the i-th point cloud sequence D i The total number of point clouds; each time index j corresponds to a sampling time within time period T.

[0061] In particular, the point cloud timing of different millimeter-wave radars cannot be guaranteed to be aligned.

[0062] Step 1402: Establish coordinate system mapping rules f based on the radar coordinate system to the target coordinate system of each millimeter-wave radar. R→V , each point cloud sequence The coordinates of all points in the system are uniformly converted into three-dimensional coordinates in the target coordinate system;

[0063] Step 1403: Use the point cloud target detection model to analyze each point cloud sequence D. i The target trajectory is predicted to obtain the corresponding first target trajectory set;

[0064] Step 1403 includes the following:

[0065] First, divide each point cloud sequence As the current point cloud sequence;

[0066] Secondly, the first frame point cloud of the current point cloud sequence is... Input a point cloud object detection model to obtain the corresponding set of object detection boxes. And based on the target matching algorithm, according to N of the current point cloud sequence i A set of target detection boxes Target matching processing yields multiple target bounding box sequences S; where each target detection box set... Includes multiple object detection boxes 1≤k≤N j N j For the set of object detection boxes The total number of detection boxes; each target box sequence S corresponds to one real target; each target box sequence S includes multiple target boxes s; each target box s includes sampling time, center point coordinates, 3D size, orientation angle and target type; the target type of all target boxes s in each target box sequence S is consistent;

[0067] Finally, each target box s in each target box sequence S is taken as a corresponding first trajectory point; and all the first trajectory points corresponding to each target box sequence S are sorted in chronological order to form a corresponding first target trajectory; and all the first target trajectories corresponding to the current point cloud sequence are combined to form a corresponding first target trajectory set.

[0068] Among them, the first target trajectory set and the point cloud sequence D i Each target trajectory is a one-to-one correspondence, consisting of multiple first target trajectories; each first target trajectory is composed of multiple first trajectory points arranged in sequence; each first trajectory point includes sampling time, center point coordinates, three-dimensional dimensions, orientation angle, and target type; the target type of all first trajectory points in each first target trajectory remains consistent.

[0069] See Figure 5 After the first point cloud sequence is input into the point cloud target detection model, a target detection sequence is obtained. The target detection sequence includes a set of target boxes, which includes target boxes. Each target box includes time, center point coordinates, length, width, height, orientation, target type, and confidence level. Then, a target matching algorithm is used to obtain a target set, which includes a sequence of target boxes. The target set includes a sequence of target boxes. Finally, target trajectory prediction is performed to obtain a target trajectory set, which includes target trajectories. Each target trajectory includes trajectory points, which include time, coordinates, and confidence level.

[0070] See Figure 6 For multiple target trajectories of the same target obtained from multiple radars, time point alignment is performed based on interpolation. The coordinates at the same time point are weighted by normalized confidence and all coordinates are weighted and summed to obtain fused coordinates. The fused target trajectory is then obtained based on the fused coordinates.

[0071] The following explains how to train a point cloud object detection model.

[0072] When selecting a model, the prototypes of the point cloud object detection model can be: PointNet series models and RDNet series models.

[0073] In terms of model functionality, the point cloud object detection model is used to perform object detection processing on the first frame of input point cloud and output multiple corresponding object detection boxes.

[0074] The first frame of the point cloud includes multiple first points; each first point includes a timestamp, grid coordinates (x, y, z), reflection intensity, radial velocity, and dynamic attributes. Note that the timestamps of each point in each frame are aligned with the frame time of the current frame, and there are no discrepancies in the timestamps of different points within the same frame. Each target detection box corresponds to a detected target; each target detection box includes: the center coordinates of the target box (aligned with a grid coordinate), the three-dimensional dimensions of the target box (length, width, height), the orientation angle of the target box, and the target type (such as people, animals, various static objects, etc.).

[0075] Model training includes determining the dataset and the loss function, and then training the model based on the dataset and the loss function, as follows:

[0076] For the dataset:

[0077] The first dataset was constructed through data collection;

[0078] The first dataset includes multiple first data records; each first data record includes a first training point cloud and a set of labeled detection boxes.

[0079] The first training point cloud is a single frame point cloud from a millimeter-wave radar, and its data format is consistent with that of the first frame point cloud; the label detection box set includes multiple label detection boxes; the data format of the label detection boxes is consistent with that of the target detection boxes.

[0080] Regarding the loss function:

[0081] Loss function L total As shown below:

[0082]

[0083] in, These are the coordinates of the center point of the predicted / label detection box; These are the three-dimensional dimensions of the predicted / label detection bounding boxes; These are the orientation angles of the predicted / label detection boxes, respectively; These represent the target types of the predicted / labeled detection boxes; L pos L size L angle L cls These are the position loss, size loss, orientation angle loss, and classification loss, respectively; position loss L pos Dimensional loss L sizeImplemented based on the Smooth L1 loss function; classification loss L... cls Implemented based on the cross-entropy loss function; n is the training sample index, 1≤n≤N tr N tr The total number of training samples is given, and each training sample corresponds one-to-one with the first data record; q is the index of the detection box, 1≤n≤N. n N n This represents the total number of detection boxes in the set of label detection boxes corresponding to the nth training sample.

[0084] The training process is as follows:

[0085] Step 1: Use the total number of records in the first dataset as the corresponding total number of training samples N. tr .

[0086] Step 2: Take each of the first data records in the first dataset as the current training record r. n ; and the current training record r n The set of label detection boxes is taken as the current label set; and the total number of detection boxes in the current label set is recorded as the corresponding total number of detection boxes N. n ; and the current training record r n The first training point cloud is input into the point cloud target detection model and processed to obtain multiple target detection boxes. Each target detection box obtained in this process is marked as an unused box. Each label detection box in the current label set is taken as the corresponding current label box. The center point coordinates, 3D dimensions, orientation angle, and target type of the current label box are recorded as the corresponding values. It then identifies whether the total number of unused bounding boxes is 0; if the total number of unused bounding boxes is not 0, it selects the unused bounding box whose center point coordinates are closest to the center point coordinates of the current label box as the predicted label box, and records the center point coordinates, 3D dimensions, orientation angle, and target type of the predicted label box as the corresponding... The currently predicted label box is changed from an unused box to a used box; if the total number of unused boxes is 0, a set of corresponding center point coordinates is set based on a set of random data. Three-dimensional dimensions Orientation Angle Target type And a set corresponding to the current label box This forms a corresponding label-prediction array.

[0087] Step 3, obtain all label-prediction arrays Input loss function L total The corresponding first loss value is obtained through calculation.

[0088] Step 4: Identify whether the first loss value meets the preset first loss value range; if not, then based on the preset first model optimizer, move towards making the loss function L... total The direction that reaches the minimum value modulates the model parameters of the point cloud target detection model in one round, and returns to step 2 after this round of modulation; if the condition is met, training stops.

[0089] Step 1404: Set a subdivided time axis L for time period T, and set multiple subdivided time points t on the subdivided time axis L based on a preset time interval Δt; and ensure that all sampling times of all first target trajectory sets can be aligned with a subdivided time point t on the subdivided time axis L.

[0090] Step 1405: Based on the subdivided time axis L, perform time alignment processing on each set of first target trajectories to obtain the corresponding set of second target trajectories;

[0091] Step 1405 includes the following:

[0092] First, the sets of trajectories of each first target are taken as the current set;

[0093] Secondly, each first target trajectory in the current set is taken as the current trajectory; and the second trajectory points corresponding to each subdivision time point t on the subdivision time axis L are set according to the interpolation method; and a corresponding second target trajectory is formed by all the second trajectory points corresponding to the current trajectory.

[0094] Each second target trajectory is composed of multiple second trajectory points arranged in sequence; each second trajectory point includes a subdivision time point, center point coordinates, three-dimensional dimensions, orientation angle, and target type; the target type of all second trajectory points in each second target trajectory is consistent; the total number of trajectory points in each second target trajectory is consistent with the total number of subdivision time points in the subdivision time axis L.

[0095] Finally, a corresponding set of second target trajectories is formed by all the second target trajectories corresponding to the current set.

[0096] Among them, the second target trajectory set and point cloud sequence D i It corresponds one-to-one and consists of multiple second target trajectories.

[0097] Step 1406: Merge the M sets of second target trajectories to obtain a first merged set; wherein, the first merged set contains all the second target trajectories of the M sets of second target trajectories;

[0098] Step 1407: Calculate the average distance between every two second target trajectories in the first merged set;

[0099] Step 1407 includes: calculating the straight-line distance between the coordinates of the two center points of each pair of second target trajectories at each subdivided time point; and calculating the average distance of the corresponding trajectory by averaging all the obtained straight-line distances.

[0100] Step 1408: Cluster all second target trajectories in the first merged set based on the average trajectory spacing to obtain multiple first trajectory clusters; wherein each first trajectory cluster consists of one or more second target trajectories; the average trajectory spacing between any two second target trajectories in each first trajectory cluster with a total number of trajectories greater than 1 is less than a preset average spacing threshold.

[0101] Step 1409: Merge the trajectories based on each first trajectory cluster to obtain a corresponding first predicted trajectory.

[0102] Step 1409 includes the following:

[0103] First, each first trajectory cluster is taken as the current trajectory cluster;

[0104] Secondly, the types of target types in the current trajectory cluster are identified, the number of each target type is counted, and the target type with the largest number is taken as the corresponding current target type; and the second target trajectory in the current trajectory cluster that does not match the current target type is deleted.

[0105] Finally, each subdivided time point t on the subdivided time axis L is taken as the current time point; and the second trajectory points corresponding to the current time point in the remaining second target trajectories are recorded as the corresponding matching trajectory points; the average of the center point coordinates of all matching trajectory points is used to calculate the corresponding current center point coordinates, the average of the three-dimensional dimensions of all matching trajectory points is used to calculate the corresponding current three-dimensional dimensions, and the average of the orientation angles of all matching trajectory points is used to calculate the corresponding current orientation angle; and the current time point and its corresponding current center point coordinates, current three-dimensional dimensions, current orientation angle, and current target type are used as a set of corresponding first time point, first center point coordinates, first three-dimensional dimensions, first orientation angle, and first target type to form a corresponding first predicted trajectory point; and all the first predicted trajectory points corresponding to the subdivided time axis L form a corresponding first predicted trajectory.

[0106] The first predicted trajectory corresponds to a real target within the actual regulatory space; the first predicted trajectory is the activity trajectory of the real target within the regulatory space during time period T; the first predicted trajectory is formed by sequentially sorting multiple first predicted trajectory points; each first predicted trajectory point includes a first time point, first center point coordinates, first three-dimensional dimensions, first orientation angle, and first target type; the total number of trajectory points of each first predicted trajectory is consistent with the total number of subdivided time points of the subdivided time axis L.

[0107] Step 150: Based on the U temperature measurement sequences, predict the temperature measurement trajectory, peak temperature, and ambient temperature difference to obtain the corresponding U first temperature measurement trajectories and a set of time period peak temperatures and time period ambient temperature differences;

[0108] The first temperature measurement trajectory includes multiple first temperature measurement trajectory points; each first temperature measurement trajectory point includes a first temperature measurement time point, a first temperature measurement coordinate, and a first temperature; the total number of trajectory points of each first temperature measurement trajectory is consistent with the total number of subdivided time points of the subdivided time axis.

[0109] Specifically, such as Figure 7 As shown, step 150 includes the following:

[0110] Step 1501: Obtain U temperature measurement sequences from U infrared sensors over time period T. 1 ≤ sensor index g ≤ U; each temperature measurement sequence Includes multiple temperature measurement data 1≤time index h≤N g N g For the g-th temperature measurement sequence W g The total number of data points; each time index h corresponds to a sampling time within time period T; each temperature measurement data point This is a temperature data set in degrees Celsius.

[0111] One issue is that the temperature measurement times of different infrared sensors cannot be guaranteed to be aligned.

[0112] Step 1502: Based on the correspondence between the installation positions of each infrared sensor and the three-dimensional coordinates of the target coordinate system, generate each temperature measurement sequence. Set a corresponding temperature measurement coordinate;

[0113] See Figure 8 Based on the temperature measurement coordinates, the first temperature lattice can be obtained, and then the temperature distribution of three-dimensional spatial points can be obtained based on the interpolation method.

[0114] Step 1503: Set a subdivided time axis L for time period T, and set multiple subdivided time points t on the subdivided time axis L based on a preset time interval Δt; and ensure that all sampling times of all first target trajectory sets can be aligned with a subdivided time point t on the subdivided time axis L.

[0115] Step 1504, analyze each temperature measurement sequence based on the subdivided time axis L. Time alignment processing is performed to obtain the corresponding first temperature measurement trajectory;

[0116] Step 1504 includes the following:

[0117] Each temperature measurement sequence As the current sequence;

[0118] Using interpolation, the temperature of each subdivided time point t on the subdivided time axis L of the current sequence is set; and each subdivided time point t and its corresponding temperature measurement coordinates and temperature are used as a set of first temperature measurement time points, first temperature measurement coordinates and first temperatures to form a corresponding first temperature measurement trajectory point; and all the first temperature measurement trajectory points corresponding to the current sequence are used to form a corresponding first temperature measurement trajectory.

[0119] The first temperature measurement trajectory includes multiple first temperature measurement trajectory points; each first temperature measurement trajectory point includes a first temperature measurement time point, a first temperature measurement coordinate, and a first temperature; the total number of trajectory points of each first temperature measurement trajectory is consistent with the total number of subdivided time points of the subdivided time axis L.

[0120] Step 1505: Based on the obtained U first temperature measurement trajectories, identify the peak temperature and ambient temperature difference of time period T.

[0121] Step 1505 includes the following:

[0122] Each subdivided time point t on the subdivided time axis L is taken as the current time point; the U first temperatures corresponding to the current time point in the U first temperature measurement trajectories are taken as the corresponding U sample temperatures; the maximum value of the U sample temperatures is taken as the single-point peak temperature of the current time point; the mean μ and standard deviation σ of the U-1 sample temperatures excluding the single-point peak temperature are calculated, and the single-point ambient temperature of the current time point is set to μ + 3σ based on the mean μ and standard deviation σ.

[0123] The maximum single-point peak temperature corresponding to the current subdivided time axis L is taken as the peak temperature of the corresponding time period.

[0124] The temperature difference between the maximum single-point peak temperature and the minimum single-point ambient temperature corresponding to the current subdivided time axis L is taken as the corresponding time period ambient temperature difference = maximum single-point peak temperature - minimum single-point ambient temperature.

[0125] Step 160: Identify and issue warnings for abnormal events related to the dwell time of personnel based on all the first predicted trajectories received this time;

[0126] Among them, see Figure 9 Step 160 includes the following:

[0127] Step 1601: Take the first predicted trajectory of each target type as the corresponding first person trajectory.

[0128] Step 1602: Take each first person's trajectory as the corresponding current trajectory; and cluster the first predicted trajectory points of the current trajectory based on the coordinates of the first center point to obtain one or more corresponding first trajectory point clusters; and take the product of the total number of trajectory points in each first trajectory point cluster and the time interval Δt as the dwell time corresponding to the current trajectory point cluster; and identify whether the maximum dwell time exceeds the preset dwell time threshold; if it exceeds, set the corresponding first person's status to abnormal; if it does not exceed, set the corresponding first person's status to normal.

[0129] Each first trajectory point cluster consists of one or more first predicted trajectory points; the straight-line distance between the center points of any two first predicted trajectory points in a first trajectory point cluster with a total number of trajectory points greater than 1 does not exceed a preset straight-line distance threshold.

[0130] Step 1603: Identify whether all the obtained first personnel statuses are normal; if at least one first personnel status is abnormal, send the corresponding personnel stay time abnormal event warning information to the preset warning interface of the regulatory space management personnel.

[0131] The warning interfaces for space management personnel include SMS, voice, email, and instant messaging software interfaces.

[0132] Step 170: Identify and issue early warnings for single-point high-temperature anomalies based on the first temperature measurement trajectories obtained this time;

[0133] For details, see Figure 10 Step 170 includes the following:

[0134] Step 1701: Take each first temperature measurement trajectory as the corresponding current temperature measurement trajectory; and take the sensor temperature difference threshold corresponding to the current temperature measurement trajectory as the current temperature difference threshold.

[0135] Step 1702: Take the highest first temperature in the current temperature measurement trajectory as the first position peak temperature, and take the average value μ of all first temperatures in the current temperature measurement trajectory excluding the first position peak temperature. * and standard deviation σ * Perform calculations based on the mean μ. * and standard deviation σ * Set the corresponding first position reference temperature = μ * +3σ * The temperature difference between the peak temperature at the first position and the reference temperature at the first position is taken as the corresponding temperature difference at the first position = peak temperature at the first position - reference temperature at the first position.

[0136] Step 1703: Identify whether the temperature difference at the first position exceeds the current temperature difference threshold; if yes, set the corresponding first position status to abnormal; if no, set the corresponding first position status to normal.

[0137] Step 1704: Identify whether all the obtained first position states are normal; if at least one first position state is abnormal, send the corresponding single-point high temperature abnormality event warning information to the warning interface of the monitoring space management personnel.

[0138] Step 180: Identify and issue early warnings for abnormal high-temperature events based on the peak temperature and ambient temperature difference obtained during this period.

[0139] Specifically, step 180 includes the following:

[0140] The system identifies peak temperature and ambient temperature difference within a given time period. If the peak temperature exceeds a preset warning temperature threshold or the ambient temperature difference exceeds a preset warning temperature difference threshold, the corresponding first environmental state is set to abnormal. If the peak temperature does not exceed the warning temperature threshold and the ambient temperature difference does not exceed the warning temperature difference threshold, the corresponding first environmental state is set to normal.

[0141] If the first environmental state is abnormal, then send the corresponding environmental high temperature abnormality event warning information to the warning interface of the monitored space management personnel.

[0142] The method for anomaly monitoring based on radar and infrared sensor data provided by this invention has the following technical effects:

[0143] Multi-source data fusion enhances monitoring accuracy and reliability. By combining point cloud data from millimeter-wave radar (providing the target's three-dimensional position, size, orientation, and trajectory) with temperature data from infrared sensors (providing temperature distribution and dynamic changes), multi-dimensional perception of targets within the monitored space is achieved. Compared to single-sensor solutions, this method effectively avoids monitoring failures caused by misjudgments or obstructions from a single data source, significantly improving the accuracy and robustness of anomaly identification.

[0144] A spatiotemporal alignment mechanism ensures collaborative data analysis. By establishing a three-dimensional grid network as a unified target coordinate system and designing mapping rules from the radar coordinate system to the target coordinate system, spatial alignment of data from multiple radars and infrared sensors is achieved. Simultaneously, by subdividing the time axis and using time point alignment strategies, the synchronization of point cloud sequences and temperature measurement sequences in the time dimension is ensured. This mechanism provides a precise spatiotemporal reference for subsequent trajectory prediction and anomaly event correlation analysis, avoiding false alarms or missed detections caused by data misalignment.

[0145] Dynamic trajectory prediction enables refined anomaly identification. Target trajectory prediction based on point cloud sequences can generate a first predicted trajectory containing parameters such as time point, grid coordinates, and 3D dimensions, accurately depicting the real-time activity status of targets within the monitored space. Combining the peak temperature of the temperature measurement trajectory with the analysis of the environmental temperature difference, this method can distinguish different types of abnormal events, such as personnel lingering (based on trajectory continuity), single-point high temperature (based on local temperature abrupt changes), and environmental high temperature (based on global temperature difference thresholds), achieving refined identification from "existence of an anomaly" to "anomaly type."

[0146] A tiered early warning mechanism improves emergency response efficiency. This method sets up independent identification and early warning logic for three types of abnormal events: personnel confinement, single-point high temperatures, and ambient high temperatures, forming a tiered response system. For example, abnormal personnel confinement time can be triggered by the duration of trajectory points; single-point high temperature anomalies can be judged by combining sensor temperature difference thresholds and temperature rise rates; and ambient high temperature anomalies are triggered by a linkage between peak temperature over a time period and dynamic ambient temperature difference thresholds. This tiered strategy prioritizes high-risk events and optimizes resource allocation efficiency.

[0147] Adaptive parameter configuration enhances scenario adaptability. By setting adjustable parameters such as the sensor temperature difference threshold for the infrared sensor and configuring dynamic time intervals for subdivided time axes, this method can adapt to the monitoring needs of different regulatory scenarios (such as indoor / outdoor, dense / sparse spaces). For example, in densely populated scenarios, the time interval can be shortened to increase the trajectory prediction frequency, and in high-temperature industrial environments, the low temperature difference threshold can be adjusted to improve sensitivity to environmental anomalies, thereby balancing monitoring accuracy and computational resource consumption.

[0148] Low-latency real-time monitoring meets safety control requirements. Each sensor independently collects data at its inherent frequency, and the data is periodically aggregated and analyzed at a preset time frequency, ensuring data integrity while reducing the real-time processing pressure on the system. A detailed time axis and alignment mechanism ensure that the latency of abnormal event identification is controllable (depending only on the time frequency setting), meeting the stringent real-time monitoring requirements of scenarios such as public safety and industrial production.

[0149] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0150] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0151] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring abnormal events based on radar and infrared sensor data, characterized in that, The method includes: M millimeter-wave radars and U infrared sensors are deployed within the monitored space; a corresponding three-dimensional mesh network is set up for the monitored space; the three-dimensional coordinate system of the three-dimensional mesh network is used as the target coordinate system; based on the radar parameters of each millimeter-wave radar, a coordinate system mapping rule from the radar coordinate system to the target coordinate system is set; based on the installation position of each infrared sensor, a corresponding mesh coordinate is set for each infrared sensor in the target coordinate system as the temperature measurement coordinate; and a corresponding sensor temperature difference threshold is set for each infrared sensor. Each millimeter-wave radar periodically scans the monitored space according to its own scanning frequency, generating and saving corresponding single-frame point clouds; each infrared sensor periodically monitors the temperature at its respective installation location according to its own temperature measurement frequency, generating and saving corresponding temperature measurement data. At a preset first time frequency, M point cloud sequences generated by M millimeter-wave radars and U temperature measurement sequences generated by U infrared sensors are periodically acquired within the most recent time period T; a corresponding subdivided time axis is set for the current time period T; and multiple subdivided time points are set on the subdivided time axis based on preset time intervals; and it is ensured that each sampling time of all point cloud sequences and each temperature measurement time of all temperature measurement sequences can be aligned with a subdivided time point on the subdivided time axis. Multiple first predicted trajectories are obtained by predicting the target trajectory based on M point cloud sequences. Each first predicted trajectory corresponds to a real target in the actual monitoring space. The first predicted trajectory is the activity trajectory of the real target in the time period and monitoring space. The first predicted trajectory is formed by sequentially sorting multiple first predicted trajectory points. Each first predicted trajectory point includes a first time point, first center point coordinates, first three-dimensional size, first orientation angle, and first target type. The first time point corresponds one-to-one with the subdivided time points. The first center point coordinates are three-dimensional grid coordinates in a three-dimensional grid network. Based on U temperature measurement sequences, the corresponding U first temperature measurement trajectories, peak temperatures, and ambient temperature differences are predicted to obtain a set of peak temperatures and ambient temperature differences for each time period. Among them, the first temperature measurement trajectory includes multiple first temperature measurement trajectory points. Each first temperature measurement trajectory point includes a first temperature measurement time point, a first temperature measurement coordinate, and a first temperature. The total number of trajectory points of each first temperature measurement trajectory is consistent with the total number of subdivided time points of the subdivided time axis. Based on all the first predicted trajectories obtained this time, we will identify and issue early warnings for abnormal events related to personnel stay time. Based on the first temperature measurement trajectories obtained this time, single-point high temperature anomaly events are identified and early warnings are issued; Based on the peak temperature and ambient temperature difference obtained during this period, we will identify and issue early warnings for abnormal high-temperature events. Specifically, the step of predicting the target trajectory based on M point cloud sequences to obtain multiple corresponding first predicted trajectories includes: Obtain M point cloud sequences from M millimeter-wave radars over time period T. Where 1 ≤ radar index i ≤ M; each point cloud sequence D i Including multiple first-frame point clouds 1 ≤ time index j ≤ N i N i Let D be the i-th point cloud sequence. i The total number of point clouds; each time index j corresponds to a sampling time within time period T; Establish coordinate system mapping rules from the radar coordinate system to the target coordinate system based on each millimeter-wave radar. R→V , each point cloud sequence The coordinates of all points in the system are uniformly converted into three-dimensional coordinates in the target coordinate system; The point cloud object detection model was used to analyze each point cloud sequence D. i The target trajectory is predicted to obtain the corresponding first target trajectory set; Set a subdivided time axis L for time period T, and set multiple subdivided time points t on the subdivided time axis L based on a preset time interval Δt; and ensure that all sampling times of all first target trajectory sets can be aligned with a subdivided time point t on the subdivided time axis L; Based on the subdivided time axis L, time alignment processing is performed on each set of first target trajectories to obtain the corresponding set of second target trajectories; The M sets of second target trajectories are merged to obtain a first merged set; wherein, the first merged set contains all the second target trajectories of the M sets of second target trajectories; The average distance between every two second target trajectories in the first merged set is calculated, including: calculating the straight-line distance between the coordinates of the two center points of every two second target trajectories at each subdivided time point; and calculating the average distance between all the obtained straight-line distances to obtain the corresponding average distance between trajectories. Based on the average trajectory spacing, all second target trajectories in the first merged set are clustered to obtain multiple first trajectory clusters; wherein each first trajectory cluster consists of one or more second target trajectories; the average trajectory spacing between any two second target trajectories in each first trajectory cluster with a total number of trajectories greater than 1 is less than a preset average spacing threshold. A first predicted trajectory is obtained by merging the trajectories of each first trajectory cluster.

2. The method according to claim 1, characterized in that, The point cloud target detection model is used to analyze each point cloud sequence D. i The target trajectory prediction process yields the corresponding first set of target trajectories, specifically: Each point cloud sequence As the current point cloud sequence; The first frame of each point cloud in the current point cloud sequence Input a point cloud object detection model to obtain the corresponding set of object detection boxes. ; And based on the Hungarian target matching algorithm, according to N of the current point cloud sequence i A set of target detection boxes Target matching processing yields multiple target bounding box sequences S; where each target detection box set... Includes multiple object detection boxes ; 1≤k≤N j N j For the set of object detection boxes The total number of detection boxes; each target box sequence S corresponds to one real target; each target box sequence S includes multiple target boxes s; each target box s includes sampling time, center point coordinates, 3D size, orientation angle and target type; the target type of all target boxes s in each target box sequence S is consistent; Each target box s in each target box sequence S is taken as a corresponding first trajectory point; all first trajectory points corresponding to each target box sequence S are sorted in chronological order to form a corresponding first target trajectory; and all first target trajectories corresponding to the current point cloud sequence are combined to form a corresponding first target trajectory set; wherein, the first target trajectory set and the point cloud sequence D are considered together. i Each target trajectory is a one-to-one correspondence, consisting of multiple first target trajectories; each first target trajectory is composed of multiple first trajectory points arranged in sequence; each first trajectory point includes sampling time, center point coordinates, three-dimensional dimensions, orientation angle, and target type; the target type of all first trajectory points in each first target trajectory remains consistent.

3. The method according to claim 1, characterized in that, The step of performing time alignment processing on each first target trajectory set based on the subdivided time axis L to obtain the corresponding second target trajectory set specifically includes: Use each set of first target trajectories as the current set; Each first target trajectory in the current set is taken as the current trajectory; and the second trajectory points corresponding to each subdivision time point t on the subdivision time axis L are set according to the interpolation method; and a corresponding second target trajectory is formed by all the second trajectory points corresponding to the current trajectory; wherein, each second target trajectory is composed of multiple second trajectory points ordered in sequence; each second trajectory point includes the subdivision time point, center point coordinates, three-dimensional dimensions, orientation angle and target type; the target type of all second trajectory points in each second target trajectory is consistent; the total number of trajectory points in each second target trajectory is consistent with the total number of subdivision time points on the subdivision time axis L; A corresponding set of second target trajectories is formed by all the second target trajectories corresponding to the current set; the set of second target trajectories and the point cloud sequence D i It corresponds one-to-one and consists of multiple second target trajectories.

4. The method according to claim 1, characterized in that, The step of merging trajectories based on each first trajectory cluster to obtain a corresponding first predicted trajectory specifically includes: Each of the first trajectory clusters is taken as the current trajectory cluster; The target types in the current trajectory cluster are identified, and the number of each target type is counted. The target type with the largest number is taken as the current target type. Second target trajectories in the current trajectory cluster whose target type does not match the current target type are deleted. Each subdivided time point t on the subdivided time axis L is taken as the current time point; the second trajectory points corresponding to the current time point in the remaining second target trajectories are recorded as the corresponding matching trajectory points; the average of the center point coordinates of all matching trajectory points is used to calculate the corresponding current center point coordinates, the average of the three-dimensional dimensions of all matching trajectory points is used to calculate the corresponding current three-dimensional dimensions, and the average of the orientation angles of all matching trajectory points is used to calculate the corresponding current orientation angle; and the current time point and its corresponding current center point coordinates, current three-dimensional dimensions, current orientation angle, and current target type are used as a set of corresponding first time points, first center point coordinates, and first three-dimensional dimensions. A first predicted trajectory point is formed by the first three-dimensional dimension, the first orientation angle, and the first target type; and a first predicted trajectory is formed by all the first predicted trajectory points corresponding to the subdivided time axis L; wherein, the first predicted trajectory corresponds to a real target in the real regulatory space; the first predicted trajectory is the activity trajectory of the real target in the time period T and the regulatory space; the first predicted trajectory is formed by sequentially sorting multiple first predicted trajectory points; each first predicted trajectory point includes a first time point, the coordinates of a first center point, a first three-dimensional dimension, a first orientation angle, and a first target type; the total number of trajectory points of each first predicted trajectory is consistent with the total number of subdivided time points of the subdivided time axis L.

5. The method according to claim 1, characterized in that, The step of predicting the temperature trajectory, peak temperature, and ambient temperature difference based on U temperature measurement sequences to obtain the corresponding U first temperature measurement trajectories and a set of time-period peak temperatures and time-period ambient temperature differences specifically includes: Obtain U temperature measurement sequences from U infrared sensors over time period T. ; 1 ≤ sensor index g ≤ U; each temperature measurement sequence Includes multiple temperature measurement data 1 ≤ time index h ≤ N g N g For the g-th temperature measurement sequence W g The total number of data points; each time index h corresponds to a sampling time within time period T; each temperature measurement data point This is a temperature data set in degrees Celsius. Based on the correspondence between the installation positions of each infrared sensor and the three-dimensional coordinates of the target coordinate system, each temperature measurement sequence is generated. Set a corresponding temperature measurement coordinate; Set a subdivided time axis L for time period T, and set multiple subdivided time points t on the subdivided time axis L based on a preset time interval Δt; and ensure that all sampling times of all first target trajectory sets can be aligned with a subdivided time point t on the subdivided time axis L; Based on the subdivided time axis L, each temperature measurement sequence Time alignment processing is performed to obtain the corresponding first temperature measurement trajectory; Based on the obtained U first temperature measurement trajectories, the peak temperature and ambient temperature difference of time period T are identified.

6. The method according to claim 5, characterized in that, The identification of the peak temperature and ambient temperature difference for time period T based on the obtained U first temperature measurement trajectories specifically includes: Each subdivided time point t on the subdivided time axis L is taken as the current time point; the U first temperatures corresponding to the current time point in the U first temperature measurement trajectories are taken as the corresponding U sample temperatures; the maximum value of the U sample temperatures is taken as the single-point peak temperature of the current time point; the mean µ and standard deviation σ of the U-1 sample temperatures excluding the single-point peak temperature are calculated, and the single-point ambient temperature of the current time point is set to µ + 3σ based on the mean µ and standard deviation σ. The maximum single-point peak temperature corresponding to the current subdivided time axis L is taken as the peak temperature of the corresponding time period. The temperature difference between the maximum single-point peak temperature and the minimum single-point ambient temperature corresponding to the current subdivided time axis L is taken as the corresponding time period ambient temperature difference = maximum single-point peak temperature - minimum single-point ambient temperature.

7. The method according to claim 1, characterized in that, The process of identifying and issuing early warnings for abnormal personnel stay times based on all the first predicted trajectories received this time specifically includes: The first predicted trajectory for each target type is a person, which is then used as the corresponding first person trajectory. Each first person's trajectory is taken as the corresponding current trajectory; and the first predicted trajectory points of the current trajectory are clustered based on the coordinates of the first center point to obtain one or more corresponding first trajectory point clusters; the product of the total number of trajectory points in each first trajectory point cluster and the time interval is taken as the dwell time corresponding to the current trajectory point cluster; and it is identified whether the maximum dwell time exceeds the preset dwell time threshold; if it exceeds, the corresponding first person's status is set to abnormal; if it does not exceed, the corresponding first person's status is set to normal; wherein, each first trajectory point cluster consists of one or more first predicted trajectory points; the straight-line distance between the center point coordinates of any two first predicted trajectory points in each first trajectory point cluster with a total number of trajectory points greater than 1 does not exceed the preset straight-line distance threshold; The system identifies whether all the first personnel statuses are normal; if at least one first personnel status is abnormal, it sends the corresponding personnel stay time abnormal event warning information to the preset warning interface of the regulatory space management personnel; the warning interface of the regulatory space management personnel includes SMS, voice, email and instant messaging software interfaces.

8. The method according to claim 1, characterized in that, The process of identifying and issuing early warnings for single-point high-temperature anomalies based on the first temperature measurement trajectories obtained this time specifically includes: Each first temperature measurement trajectory is taken as the corresponding current temperature measurement trajectory; and the sensor temperature difference threshold corresponding to the current temperature measurement trajectory is taken as the current temperature difference threshold. The highest first temperature in the current temperature measurement trajectory is taken as the first position peak temperature, and the average value µ of all first temperatures in the current temperature measurement trajectory excluding the first position peak temperature is taken. * and standard deviation σ * Perform calculations based on the mean µ * and standard deviation σ * Set the corresponding first position reference temperature = µ * +3σ * The temperature difference between the peak temperature at the first position and the reference temperature at the first position is taken as the corresponding temperature difference at the first position = peak temperature at the first position - reference temperature at the first position; The system identifies whether the temperature difference at the first location exceeds the current temperature difference threshold; if so, it sets the corresponding first location status to abnormal; otherwise, it sets the corresponding first location status to normal. The system identifies whether all obtained first position statuses are normal; if at least one first position status is abnormal, it sends the corresponding single-point high temperature anomaly warning information to the warning interface of the monitoring space management personnel.

9. The method according to claim 1, characterized in that, The identification and early warning of abnormal high-temperature events based on the peak temperature and ambient temperature difference obtained during this period specifically includes: The system identifies peak temperature and ambient temperature difference within a given time period. If the peak temperature exceeds a preset warning temperature threshold or the ambient temperature difference exceeds a preset warning temperature difference threshold, the corresponding first environmental state is set to abnormal. If the peak temperature does not exceed the warning temperature threshold and the ambient temperature difference does not exceed the warning temperature difference threshold, the corresponding first environmental state is set to normal. If the first environmental state is abnormal, then send the corresponding environmental high temperature abnormality event warning information to the warning interface of the monitored space management personnel.

Images