Dangerous behavior feature identification method for Internet of Things security of smart community

Abstract

The invention discloses a dangerous behavior feature recognition method for smart community Internet of Things security, and relates to the technical field of behavior recognition, the method comprises the steps that a camera detects movement information of personnel and vehicles in a community and analyzes dangerous behavior features of the personnel and the vehicles, and the method has the advantages that a complex scene model of the community is constructed through a 3D modeling technology; even if the target is partially shielded in some areas, the behavior of the target is analyzed more meticulously and the complete motion trail of the target is tracked accurately by using the spatial information of the scene and the posture, action and other characteristics of the remaining visible part of the target, so that whether the target has abnormal behaviors such as illegal entry into the restricted area, wandering snooping and the like is judged more accurately; even if the part on the road is shielded by other vehicles or buildings, whether traffic rule violation or potential safety hazards exist or not can be judged according to the relative position of the part on the road in the 3D map and the driving state of the exposed parts such as the head and the tail of the vehicle.

Description

Technical Field

[0001] This invention relates to the field of behavior recognition technology, specifically a method for identifying dangerous behavior characteristics in smart community IoT security. Background Technology

[0002] Visual recognition of abnormal human behavior for safe production focuses on using video surveillance systems and computer vision technology to identify behaviors or activities that may endanger workplace safety. Research in this field mainly focuses on how to automatically parse video content through algorithms to identify abnormal or dangerous behavioral patterns. Human dangerous behavior recognition is an important research area in intelligent surveillance systems, which is crucial for crime prevention. Automatically detecting and identifying dangerous human behaviors in video surveillance scenarios in public places such as train stations, stadiums, and shopping malls can promptly identify potential accidents and hazards and handle them immediately. Establishing a comprehensive and timely early warning mechanism can effectively reduce or eliminate public safety issues and ensure personal and property safety. In the real world, due to many factors such as video shooting angle, background brightness, diversity of features, and relationships between human behaviors, feature-based behavior detection has become a focus of researchers. Complex backgrounds in the scene can also increase the difficulty of recognition. If there are complex building structures, vegetation or advertising signs in the monitored area, they may interfere with the separation of the target from the background, leading to incorrect detection results. In garden areas with many trees and flower beds, the outline of a person may blend into the vegetation background, making posture analysis and behavior recognition difficult. To address this, we propose a dangerous behavior feature recognition method for smart community IoT security. Summary of the Invention

[0003] The purpose of this invention is to provide a method for identifying dangerous behavior characteristics in smart community IoT security.

[0004] To address the problems mentioned in the background section, this invention provides the following technical solution: a method for identifying dangerous behavior characteristics in smart community IoT security, comprising using cameras to detect the movement information of people and vehicles in the community, and analyzing the dangerous behavior characteristics of people and vehicles. The specific operation steps of the method for identifying dangerous behavior characteristics in smart community IoT security are as follows: Step 1: Extract two-dimensional and three-dimensional features of people and vehicles in the community using cameras and LiDAR in the IoT security system; Step 2: After integrating and labeling the collected two-dimensional and three-dimensional features of people and vehicles, use the labeled two-dimensional and three-dimensional features to train the convolutional neural network model, so that the convolutional neural network model learns the method of recognizing people and vehicle behavior. Step 3: Introduce blockchain technology into the data collection, transmission and storage process of cameras and LiDAR in the IoT security system, store the collected dangerous behavior characteristics data in the blockchain, and enable data sharing between the community and external emergency rescue departments; Step 4: Use 3D modeling technology to model complex scenes in the community. Obtain 3D structural information of the scene through cameras and LiDAR, and perform 3D map modeling. When tracking a target, if the target is partially occluded, infer its position and behavior through other visible parts and the 3D map structure of the scene. Step 5: Construct a multimodal early warning system. Based on the type and degree of dangerous behavior, adopt different combinations of early warning methods to achieve intelligent linkage between IoT security devices.

[0005] As a further aspect of the present invention: In step one, two-dimensional features of mobile personnel within the community are extracted from the camera monitoring images within the IoT security system. These two-dimensional features include the color, style, and pattern of the clothing worn by the personnel; the length, hairstyle, and color of their hair are observed; facial recognition technology is used to extract the shape, position, and size of facial features, as well as facial texture features; the personnel's body proportions, stride length, and cadence are analyzed; and the personnel's stride length and cadence are calculated using formulas.

[0006]

[0007] in, The stride of the delegates The total number of frames representing the movement trajectory of the data collector, ( , Representatives at the Frame position, The pace of the representatives The time when the representative figure finished walking. The system analyzes the three-dimensional features of a person by using multi-view video collected by cameras to determine when they begin walking. These features include the person's height data, volume and spatial morphology information of various body parts. The system reconstructs the three-dimensional outline of the person's body using the multi-view video data. Cameras installed in the community are used to capture the person's movement trajectory in three-dimensional space. Three-dimensional motion capture technology is then used to analyze the person's movements and postures, including the amplitude and direction of complex movements such as climbing, jumping and bending, as well as the changes in the position of various body parts in space.

[0008] As a further aspect of the present invention: In step one, the two-dimensional features of the vehicle are extracted from the camera monitoring images in the IoT security system, the overall color of the vehicle and the special coating on the surface of the vehicle body are identified, the license plate number is extracted by license plate recognition technology, the length, width, height and three-dimensional shape of the vehicle body are obtained by multi-view video, and the three-dimensional driving trajectory of the vehicle in the community is recorded by LiDAR, including the trajectory changes of the vehicle on flat roads and uphill and downhill road conditions.

[0009] As a further aspect of the present invention: In step two, the extracted two-dimensional and three-dimensional features of personnel and vehicles are integrated to form a feature vector set containing feature information. Based on the actual behavior observed within the community, the integrated feature data is labeled. For personnel behavior, normal walking and talking behaviors are labeled as normal, while climbing walls and loitering in restricted areas are labeled as dangerous. For vehicle behavior, normal entry and exit from the community and parking according to regulations are labeled as normal, while speeding, illegal parking, and entering pedestrian areas are labeled as dangerous. The labeled feature vector set is divided into a training set, a validation set, and a test set. The convolutional neural network model is trained using the training set, and the hyperparameters of the convolutional neural network model are adjusted using the validation set. Finally, the accuracy, recall, and F1 score of the convolutional neural network model are evaluated on the test set to determine the effectiveness of the convolutional neural network model in recognizing personnel and vehicle behavior.

[0010] As a further aspect of the present invention: In step three, during the data collection, transmission, and storage process of cameras and lidar in the IoT security system, blockchain technology is introduced to encrypt the collected dangerous behavior characteristic data, store the encrypted data in the blockchain, and enable data sharing between the community and external emergency rescue departments through the blockchain. At the same time, based on the timestamps and data source information recorded on the blockchain, the behavioral trajectories of personnel and vehicles before and after the occurrence of dangerous behavior are restored in chronological order.

[0011] As a further aspect of the present invention: In step four, a static scan of the community scene is performed using lidar installed on rooftops and utility poles to obtain point cloud data of static objects such as buildings, roads, and greenery within the community. The lidar is then mounted on a drone, which moves within the community to collect data and obtain scene information from different perspectives. The data from the camera is fused with the lidar point cloud data, and the image information obtained from the camera provides texture and color information for the point cloud data. For point cloud data collected from different locations and angles, accelerated robust features are used to extract key points and feature descriptors from the point cloud data. The correspondence between different point cloud features is found by calculating the similarity between feature descriptors. Based on the matched feature point pairs, the rotation and translation transformation matrix is ​​calculated using the least squares method to align the point cloud data to the same coordinate system.

[0012] As a further aspect of the present invention: In step four, a deep learning algorithm is used to segment and classify the preprocessed point cloud data, identify different objects and scene elements, extract feature points and feature lines from the point cloud data, construct the skeleton of a 3D map using the feature point and feature line information, and generate a complete 3D map using a formula, the calculation formula being:

[0013] in, This represents the additional offset added to the interpolation point. Represents the interpolation point. This represents the adjustment factor, with a value range of 0-1. Represents point cloud functions. The gradient of the point cloud function is used to update the map using real-time data collected by LiDAR. When new objects are added to the community or existing objects change, the scene data in the map is updated in a timely manner, and the constructed 3D map is added to the blockchain. With authorization, the collected data is uploaded to the blockchain, and the behavior of people and vehicles is identified through a convolutional neural network model.

[0014] As a further aspect of the present invention: In step five, a multimodal early warning system is constructed to classify dangerous behaviors into three categories: security threats, environmental safety, and public order. Security threats include intrusion and theft; environmental safety includes fire, gas leaks, and electrical faults; and public order includes illegal parking, improper pet management, and noise pollution. The degree of danger is classified as mild, moderate, and severe. Mild danger includes behaviors in the public order category that affect the lives of community residents; moderate danger includes situations in the security threats and environmental safety categories that have not yet caused serious consequences; and severe danger includes situations that seriously threaten the lives and property of people.

[0015] As a further aspect of the present invention: In step five, the degree of dangerous behavior is identified through a multimodal early warning system. When a minor dangerous behavior occurs, a warning is issued in green text on the monitoring screen within the community, and a notification is simultaneously pushed to the mobile phones of relevant area management personnel, informing them of the situation and location of the dangerous behavior. The community broadcast also reminds personnel and vehicles to correct the dangerous behavior. When a moderate dangerous behavior occurs, a warning is issued in flashing yellow text on the monitoring screen within the community, and a notification is simultaneously pushed to the mobile phones of all security personnel and relevant area management personnel, informing them of the situation and location of the dangerous behavior. An intermittent alarm is emitted from the audible and visual alarm at the location of the dangerous behavior. When a severe dangerous behavior occurs, a warning is issued in flashing red text on the monitoring screen within the community, and a notification is simultaneously pushed to the mobile phones of all community residents, security personnel, and relevant area management personnel, informing them of the situation and location of the dangerous behavior. An audible and visual alarm is activated throughout the community, emitting a strong and continuous alarm sound. Simultaneously, the community broadcast announces the dangerous behavior and evacuation instructions. The community IoT security system shares relevant data with the fire department via blockchain, enabling the fire department to promptly obtain information on dangerous behaviors and conduct emergency response.

[0016] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows: 1. This invention constructs a complex scene model of a community using 3D modeling technology. Even if the target is partially obscured in some areas, the spatial information of the scene and the posture, movement and other features of the remaining visible part of the target can be used to analyze the target behavior in more detail and accurately track its complete movement trajectory. This allows for a more accurate judgment of whether the target has violated regulations by entering restricted areas, loitering or spying, or other abnormal behaviors. For vehicles, even if they are partially obscured by other vehicles or buildings on the road, their relative position in the 3D map and the driving status of the exposed front and rear parts can be used to determine whether there are behaviors such as speeding, illegal lane changes, or other violations of traffic rules or potential safety hazards. 2. This invention provides a trusted sharing platform through blockchain. All data sharing operations are recorded on the blockchain, including information such as the sharing time, content, data source, and recipient. Community property management can share 3D map data with the fire department. Upon receiving a fire alarm, the fire department can quickly formulate a rescue plan based on accurate 3D map information. Different departments may have different needs and concerns when using 3D map data. Through blockchain technology, cross-departmental collaboration can be easily achieved. In community renovation projects, planning, construction, and property management departments can jointly access and update 3D map data on the blockchain. The planning department can modify future construction areas in the 3D map according to the community's development plan. The construction department can update the 3D map in real time during construction to reflect the actual construction progress. The property management department can use the updated 3D map to better arrange the layout and management of community facilities. Blockchain technology promotes information flow and collaboration efficiency among departments, enabling 3D map data to play a greater role in multiple fields. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for identifying dangerous behavior characteristics in smart community IoT security according to an embodiment of the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0019] This invention discloses a method for identifying dangerous behavior characteristics in smart community IoT security, comprising using cameras to detect the movement information of people and vehicles in the community, and analyzing the dangerous behavior characteristics of people and vehicles. The specific operation steps of the method for identifying dangerous behavior characteristics in smart community IoT security are as follows: Step 1: Extract two-dimensional and three-dimensional features of people and vehicles in the community using cameras and LiDAR in the IoT security system; Step 2: After integrating and labeling the collected two-dimensional and three-dimensional features of people and vehicles, use the labeled two-dimensional and three-dimensional features to train the convolutional neural network model, so that the convolutional neural network model learns the method of recognizing people and vehicle behavior. Step 3: Introduce blockchain technology into the data collection, transmission and storage process of cameras and LiDAR in the IoT security system, store the collected dangerous behavior characteristics data in the blockchain, and enable data sharing between the community and external emergency rescue departments; Step 4: Use 3D modeling technology to model complex scenes in the community. Obtain 3D structural information of the scene through cameras and LiDAR, and perform 3D map modeling. When tracking a target, if the target is partially occluded, infer its position and behavior through other visible parts and the 3D map structure of the scene. Step 5: Construct a multimodal early warning system. Based on the type and degree of dangerous behavior, adopt different combinations of early warning methods to achieve intelligent linkage between IoT security devices.

[0020] In one embodiment of the present invention: In step one, two-dimensional features of mobile personnel within the community are extracted from the camera monitoring images within the IoT security system. These two-dimensional features include the color, style, and pattern of the clothing worn by the personnel; the length, hairstyle, and color of the personnel's hair are observed; facial recognition technology is used to extract the shape, position, and size of facial features, as well as facial texture features; the personnel's body proportions, stride length, and cadence are analyzed; and the personnel's stride length and cadence are calculated using formulas.

[0021]

[0022] in, The stride of the delegates The total number of frames representing the movement trajectory of the data collector, ( , Representatives at the Frame position, The pace of the representatives The time when the representative figure finished walking. The system analyzes the three-dimensional features of a person by using multi-view video collected by cameras to determine when they begin walking. These features include the person's height data, volume and spatial morphology information of various body parts. The system reconstructs the three-dimensional outline of the person's body using the multi-view video data. Cameras installed in the community are used to capture the person's movement trajectory in three-dimensional space. Three-dimensional motion capture technology is then used to analyze the person's movements and postures, including the amplitude and direction of complex movements such as climbing, jumping and bending, as well as the changes in the position of various body parts in space.

[0023] In one embodiment of the present invention: in step one, the two-dimensional features of the vehicle are extracted from the camera monitoring images in the IoT security system, the overall color of the vehicle and the special coating on the surface of the vehicle body are identified, the license plate number is extracted by license plate recognition technology, the length, width, height and three-dimensional shape of the vehicle body are obtained by multi-view video, and the three-dimensional driving trajectory of the vehicle in the community is recorded by LiDAR, including the trajectory changes of the vehicle on flat roads and uphill and downhill road conditions.

[0024] In one embodiment of the present invention: In step two, the extracted two-dimensional and three-dimensional features of personnel and vehicles are integrated to form a feature vector set containing feature information. Based on the actual behavior observed in the community, the integrated feature data is labeled. For personnel behavior, normal walking and talking behaviors are labeled as normal, while climbing walls and loitering in restricted areas are labeled as dangerous. For vehicle behavior, normal entry and exit from the community and parking according to regulations are labeled as normal, while speeding, illegal parking, and entering pedestrian areas are labeled as dangerous. The labeled feature vector set is divided into a training set, a validation set, and a test set. The convolutional neural network model is trained using the training set, and the hyperparameters of the convolutional neural network model are adjusted using the validation set. Finally, the accuracy, recall, and F1 score of the convolutional neural network model are evaluated on the test set to determine the effectiveness of the convolutional neural network model in recognizing personnel and vehicle behavior.

[0025] In one embodiment of the present invention: In step three, during the data collection, transmission and storage process of cameras and lidar in the IoT security system, blockchain technology is introduced to encrypt the collected dangerous behavior characteristic data, store the encrypted data in the blockchain, and enable data sharing between the community and external emergency rescue departments through the blockchain. At the same time, based on the timestamp and data source information recorded on the blockchain, the behavioral trajectories of personnel and vehicles before and after the occurrence of dangerous behavior are restored in chronological order.

[0026] In one embodiment of the present invention: In step four, a static scan of the community scene is performed by a LiDAR installed on the rooftop and the top of a utility pole to obtain point cloud data of static objects such as buildings, roads and greenery within the community. The LiDAR is installed on a drone and moves within the community to collect data, obtaining scene information from different perspectives. The data from the camera is fused with the LiDAR point cloud data. The image information obtained by the camera provides texture and color information for the point cloud data. For point cloud data collected from different positions and angles, accelerated robust features are used to extract key points and feature descriptors from the point cloud data. The correspondence between different point cloud features is found by calculating the similarity between feature descriptors. Based on the matched feature point pairs, the rotation and translation transformation matrix is ​​calculated using the least squares method to align the point cloud data to the same coordinate system.

[0027] In one embodiment of the present invention: In step four, a deep learning algorithm is used to segment and classify the preprocessed point cloud data, identify different objects and scene elements, extract feature points and feature lines from the point cloud data, construct the skeleton of a 3D map using the feature point and feature line information, and generate a complete 3D map using a formula, the calculation formula being:

[0028] in, This represents the additional offset added to the interpolation point. Represents the interpolation point. This represents the adjustment factor, with a value range of 0-1. Represents point cloud functions. The gradient of the point cloud function is used to update the map using real-time data collected by LiDAR. When new objects are added to the community or existing objects change, the scene data in the map is updated in a timely manner, and the constructed 3D map is added to the blockchain. With authorization, the collected data is uploaded to the blockchain, and the behavior of people and vehicles is identified through a convolutional neural network model.

[0029] In one embodiment of the present invention: In step five, a multimodal early warning system is constructed to classify dangerous behaviors into three categories: security threats, environmental safety, and public order. Security threats include intrusion and theft; environmental safety includes fire, gas leaks, and electrical faults; and public order includes illegal parking, improper management of pets, and noise pollution. The degree of danger is classified as mild, moderate, and severe. Mild danger includes behaviors in the public order category that affect the lives of community residents; moderate danger includes situations in the security threats and environmental safety categories that have not yet caused serious consequences; and severe danger includes situations that seriously threaten the lives and property of people.

[0030] In one embodiment of the present invention: In step five, the degree of dangerous behavior is identified through a multimodal early warning system. When a minor dangerous behavior occurs, a warning is issued in green text on the monitoring screen within the community, and a notification is simultaneously pushed to the mobile phones of the relevant area's management personnel, informing them of the situation and location of the dangerous behavior. The community broadcast also reminds personnel and vehicles to correct the dangerous behavior. When a moderate dangerous behavior occurs, a warning is issued in flashing yellow text on the monitoring screen within the community, and a notification is simultaneously pushed to the mobile phones of all security personnel and the relevant area's management personnel, informing them of the situation and location of the dangerous behavior. An intermittent alarm is emitted from the audible and visual alarm at the location of the dangerous behavior. When a severe dangerous behavior occurs, a warning is issued in flashing red text on the monitoring screen within the community, and a notification is simultaneously pushed to the mobile phones of all community residents, security personnel, and the relevant area's management personnel, informing them of the situation and location of the dangerous behavior. An audible and visual alarm is activated throughout the entire community, emitting a strong and continuous alarm sound. Simultaneously, the community broadcast announces the dangerous behavior and evacuation instructions. The community IoT security system shares relevant data with the fire department through blockchain, enabling the fire department to promptly obtain information on dangerous behaviors and conduct emergency response.

[0031] Example 1: In a community, cameras installed at different locations can capture the movement trajectories of people in three-dimensional space. Compared to two-dimensional trajectories, three-dimensional trajectories can more comprehensively reflect people's behavior. In multi-story parking garages or areas with staircases, elevators, or other multi-level transportation facilities, the three-dimensional movement trajectories can show whether people are abnormally moving between floors or entering unauthorized areas (such as building rooftops or basement equipment rooms). This three-dimensional trajectory information can be combined with time series analysis. If a person moves rapidly from the ground floor to the top floor of a building late at night and stays there for a long time, it may indicate dangerous behavior, such as stealing equipment or attempting to damage building facilities. Using three-dimensional motion capture technology, the movement postures of people can be analyzed more accurately. For some complex movements, such as climbing, jumping, and bending over, the analysis can be performed more precisely. 3D analysis can accurately record the amplitude and direction of movements, as well as the changes in the position of various body parts in space. In identifying dangerous behaviors, such as determining whether a person is attempting to climb over a community wall or fence, 3D motion posture analysis can provide more accurate judgment than 2D video surveillance. At the same time, through 3D analysis of multiple consecutive movements, it can also identify some potentially dangerous behavioral patterns, such as a person making a series of aggressive movements in the community's fitness equipment area (such as simulating boxing, swinging a stick, etc.). When a vehicle ID "V001" is detected, the 2D features such as color and vehicle type extracted from the 2D image, and the 3D features such as the vehicle body's 3D dimensions and component spatial positions extracted from the 3D point cloud data obtained by LiDAR, are all associated and integrated with the ID "V001" to form a complete feature set about the vehicle.

[0032] Example 2: Based on the needs of community abnormal behavior detection, a convolutional neural network (CNN) model structure is constructed. A large amount of labeled sample data of complex abnormal behaviors and normal behaviors is collected. This data can be generated based on actual cases and simulated scenarios within the community. The model labels samples of complex abnormal behaviors such as people climbing walls and vehicles parking haphazardly and blocking fire lanes, as well as corresponding normal behavior samples (such as people walking normally and vehicles parking normally). This labeled data is input into the CNN model for training. The prediction results of the CNN model are calculated through forward propagation. Then, the loss value between the prediction results and the true labels is calculated based on the loss function. The gradient of the loss value with respect to the CNN model parameters is then calculated using the backpropagation algorithm. Finally, the optimizer updates the CNN model parameters based on the gradient. By adjusting the model's hyperparameters, such as the number of layers, kernel size, stride, learning rate, batch size, and regularization parameters during training, the model can accurately distinguish different behavior categories, achieving better detection accuracy.

[0033] Example 3: When a smoke sensor or temperature sensor detects signs of fire, it immediately transmits the signal to the fire alarm controller. The fire alarm controller triggers the audible and visual alarms to sound an alarm, and simultaneously sends the fire information to the community's multimodal early warning system. The multimodal early warning system automatically switches to the monitoring video footage of the fire area and determines the specific location and extent of the fire through intelligent analysis. At the same time, it sends alarm information to the fire department, including detailed information such as the community address and the location of the fire. In conjunction with the fire protection system, it automatically activates the sprinkler system to extinguish the fire, closes doors and windows near the fire area to prevent the fire from spreading, turns on emergency lighting and evacuation indicator lights to guide residents to evacuate, and notifies the elevator control system to lower the elevator to the first floor and stop operating to prevent people from being trapped.

[0034] Example 4: Collect geological structure and historical earthquake data of the community area, analyze the earthquake risk level of different areas, and intuitively mark these risk assessment results on a 3D map with different colors or transparency, such as high-risk areas represented by red and low-risk areas represented by green, so that residents and managers can clearly understand the earthquake risk distribution of the community. Integrate an earthquake monitoring sensor network into the blockchain to obtain data such as seismic wave propagation speed and acceleration in real time. When an earthquake is detected, quickly locate the epicenter on the 3D map and display the propagation range and intensity changes of seismic waves with dynamic icons, providing real-time earthquake early warning information for community residents and emergency rescue personnel.

[0035] As attached Figure 1 As shown, two-dimensional and three-dimensional features of people and vehicles in the community are extracted, and the extracted features are used to train a convolutional neural network model. 3D modeling technology is used to model complex scenes in the community, and blockchain technology is introduced in the process of data collection, transmission and storage.

[0036] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

[0037] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0038] The above description is merely an example and illustration of the present invention. Any modifications, additions, or substitutions made by those skilled in the art to the specific embodiments described, as long as they do not deviate from the invention or exceed the scope defined in the claims, shall fall within the protection scope of the present invention.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A dangerous behavior feature identification method for smart community Internet of Things security, comprising: a camera detecting moving information of personnel and vehicles in a community; and analyzing dangerous behavior features of the personnel and vehicles when moving on a community road, characterized in that, The specific operation steps of the dangerous behavior feature recognition method of the smart community IOT security system are as follows: Step one, through the camera and laser radar in the IOT security system, the two-dimensional and three-dimensional features of the personnel and vehicles in the community are extracted; Step two, after integrating the two-dimensional and three-dimensional features of the personnel and vehicles collected, labeling is performed, and the labeled two-dimensional and three-dimensional features are used to train the convolutional neural network model, so that the convolutional neural network model learns the method of personnel and vehicle behavior recognition; Step three, in the data collection, transmission and storage process of the camera and laser radar in the IOT security system, the blockchain technology is introduced, the collected dangerous behavior feature data is stored in the blockchain, and data sharing between the community and the external emergency rescue department is realized; Step four, use 3D modeling technology to model the complex scene in the community, obtain the 3D structure information of the scene through the camera and laser radar, and perform 3D map modeling. When tracking the target, if the target is partially blocked, the position and behavior of the target can be inferred through other visible parts and the 3D map structure of the scene; Step five, build a multi-modal early warning system, according to the type and degree of dangerous behavior, use different warning mode combinations to realize the intelligent linkage between IOT security devices. 2.The dangerous behavior feature recognition method of the intelligent community Internet of Things security according to claim 1, characterized in that: In step one, the two-dimensional features of the moving personnel in the community are extracted from the camera monitoring images in the IOT security system, including the color, style and pattern of the clothing worn by the personnel, the length, style and color of the hair, the shape, position, size and texture features of the face are extracted by face recognition technology, the body proportion, step size and step frequency of the personnel are analyzed, and the step size and step frequency of the personnel are calculated by formula: ； ； wherein, the pace of the representative person, the total number of frames representing the motion trajectory of the collector, , the position of the representative person at the frame, the step frequency of the representative person, the time when the representative person finishes walking, the time when the representative person starts walking, using a multi-view video analysis to obtain the three-dimensional features of the personnel, the three-dimensional features including the height data of the personnel and the volume and spatial form information of each part of the body, reconstructing the three-dimensional contour of the body of the personnel through multi-view video data, obtaining the motion trajectory of the personnel in the three-dimensional space with the help of the camera installed in the community, using three-dimensional motion capture technology to analyze the action posture of the personnel, including the amplitude, direction and position change of each part of the body in space of complex actions such as climbing, jumping and bending. 3.The dangerous behavior feature recognition method of the intelligent community IOT security according to claim 1, characterized in that: In step one, the two-dimensional features of the vehicle are extracted from the camera monitoring images in the IOT security system, including the color of the whole vehicle and the special coating on the surface of the vehicle body, the license plate number is extracted by license plate recognition technology, the length, width and height of the vehicle are obtained by multi-view video, and the three-dimensional shape of the vehicle body is obtained, and the three-dimensional driving track of the vehicle in the community is recorded by laser radar, including the track change of the vehicle on the flat road and the uphill road.

4. The dangerous behavior feature recognition method of the intelligent community Internet of Things security according to claim 3, characterized in that: In step two, the two-dimensional and three-dimensional features of the personnel and vehicles are integrated to form a feature vector set containing feature information, according to the actual behavior observed in the community, the integrated feature data is labeled, for personnel behavior, normal walking and talking behavior is labeled as normal category, climbing the fence and wandering in the forbidden area is labeled as dangerous category, for vehicle behavior, normal entry and exit of the community, parking according to regulations is labeled as normal, speeding, illegal parking and entering pedestrian area is labeled as dangerous, the labeled feature vector set is divided into training set, validation set and test set, the convolutional neural network model is trained by using the training set, the hyperparameters of the convolutional neural network model are adjusted by using the validation set, finally the accuracy, recall rate and F1 value indicators of the convolutional neural network model are evaluated on the test set, to judge the effectiveness of the convolutional neural network model for personnel and vehicle behavior recognition. 5.The dangerous behavior feature recognition method of the intelligent community IOT security according to claim 4, characterized in that: In step three, in the process of data collection, transmission and storage of cameras and lidar in the Internet of Things security system, blockchain technology is introduced to encrypt the collected dangerous behavior feature data, store the encrypted data in the blockchain, and realize data sharing between the community and external emergency rescue departments through the blockchain. At the same time, based on the time stamp and data source information recorded on the blockchain, the behavior trajectory of personnel and vehicles before and after the dangerous behavior is restored in chronological order. 6.The method of claim 1, wherein the method further comprises: In step four, the laser radar installed on the roof and the top of the telegraph pole scans the community scene statically to obtain the point cloud data of the static objects in the community buildings, roads and greenery. The laser radar is installed on the unmanned aerial vehicle to collect data while moving in the community to obtain scene information from different perspectives. The camera data and lidar point cloud data are fused, the image information obtained by the camera provides texture and color information for the point cloud data, and for the point cloud data collected from different positions and angles, the speed robust feature is used to extract key points and feature descriptors in the point cloud data, the corresponding relationship between different point cloud features is found by calculating the similarity between the feature descriptors, and the rotation and translation transformation matrix is calculated by the least squares method according to the matched feature point pairs. The point cloud data is aligned to the same coordinate system.

7. The dangerous behavior feature recognition method of the intelligent community Internet of Things security according to claim 6, characterized in that: In step four, the preprocessed point cloud data is segmented and classified using a deep learning algorithm to identify different objects and scene elements, extract feature points and feature lines from the point cloud data, and use the feature point and feature line information to construct the skeleton of the 3D map. The complete 3D map is generated by the formula: ； wherein, represents an offset newly added to the interpolation point, represents an interpolation point, represents an adjustment coefficient, the value range is 0-1, represents a point cloud function, represents the gradient of the point cloud function, the map is updated through the real-time acquisition data of the laser radar, when new objects join in the community and the original objects change, the scene data in the map is updated in time, and the constructed 3D map is added to the blockchain, the authorized data collected is uploaded to the blockchain, and the behaviors of personnel and vehicles are identified through a convolutional neural network model. 8.The method of claim 1, wherein the method further comprises: determining a dangerous behavior of the object based on the feature information. In step five, a multi-modal early warning system is constructed to classify dangerous behaviors into safety threat, environmental safety and public order categories. Safety threat includes personnel intrusion and theft, environmental safety includes fire, gas leakage and electrical fault, and public order includes vehicle illegal parking, pet management and noise disturbance. The dangerous degree is divided into mild, moderate and severe. Mild includes behaviors that affect the life of community residents, moderate includes situations that have not caused serious consequences, and severe includes situations that seriously threaten the safety of personnel and property. 9.The method of claim 8, wherein the method further comprises: In the fifth step, the degree of dangerous behavior is identified by the multi-modal early warning system. When a mild dangerous behavior occurs, a green text prompt is displayed on the monitoring screen in the community, and a notification is pushed to the mobile phones of the relevant area managers, informing them of the dangerous behavior and location. The community broadcast reminds people and vehicles to correct the dangerous behavior. When a moderate dangerous behavior occurs, a yellow flashing text prompt is displayed on the monitoring screen in the community, and a notification is pushed to the mobile phones of all security personnel and relevant area managers, informing them of the dangerous behavior and location. An intermittent alarm sound is emitted by the sound and light alarm at the location of the dangerous behavior. When a severe dangerous behavior occurs, a red flashing text prompt is displayed on the monitoring screen in the community, and a notification is pushed to the mobile phones of all community residents, security personnel, and relevant area managers, informing them of the dangerous behavior and location. The sound and light alarm is activated throughout the community, emitting a strong and continuous alarm sound. At the same time, the community broadcast voice broadcasts the dangerous behavior and evacuation instructions. The community Internet of Things security system shares relevant data with the fire department through blockchain, allowing the fire department to obtain dangerous behavior information in a timely manner and perform emergency handling.

Images