Airport boundary intrusion alarm system based on millimeter wave radar and video analysis

Abstract

The invention belongs to the technical field of airport security and protection, and relates to an airport boundary intrusion alarm system based on millimeter-wave radar and video analysis, and the system comprises a data collection module which is used for obtaining target point cloud data and video data; the preprocessing and feature extraction module is used for preprocessing video frames of the target point cloud data and the video data and extracting radar features and image features; the multi-modal feature fusion module is used for realizing fusion of the radar features and the image features through an attention mechanism by using a Transform network so as to obtain fusion features; the intrusion target detection and recognition module is used for recognizing and positioning an intrusion target and classifying and judging intrusion behaviors; and the alarm and linkage module is used for giving an alarm and recording related data of the intrusion behavior. According to the invention, the intrusion behavior can be accurately detected in real time, the false alarm and missing alarm conditions can be reduced, the automation and intelligence level of airport boundary safety management is improved, and a solid guarantee is provided for the safe operation of an airport.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of airport security and defense, and relates to an airport perimeter intrusion alarm system, in particular to an airport perimeter intrusion alarm system based on millimeter wave radar and video analysis. BACKGROUND

[0002] The field of airport security and defense is always the top priority of airport safety operation and management, and undertakes the safety guarantee function of passengers, goods, aircraft and other aviation facilities in the airport. The airport perimeter is the most direct protective barrier between the airport and the outside world. Good safety and protection of the perimeter is equivalent to guarding the first line of defense of airport security and defense.

[0003] In the traditional civil aviation airport perimeter security and defense, the existing technology mainly adopts vibration sensors, infrared beams, passive infrared, electronic fences, video monitoring and other means. The vibration sensor judges the intrusion behavior by sensing the physical vibration of the perimeter; the infrared beam uses the transmitter and receiver to form an infrared beam, and the light beam is blocked to determine the intrusion; the passive infrared detector identifies potential threats by detecting the infrared radiation emitted by the human body; the electronic fence generates an alarm signal when the lead wire is touched or damaged by arranging lead wires with pulse voltage on the perimeter; video monitoring relies on manual real-time viewing of camera-captured images to discover abnormalities. Some more advanced systems also try to use single millimeter wave radar or video analysis technology for intrusion detection, but the overall effect still has obvious shortcomings.

[0004] These single sensor security technologies have many problems, and these problems are caused by their own technical principles and are difficult to eradicate. For example, vibration sensors, infrared beams, passive infrared, and electronic fence systems are easily affected by complex weather and surrounding environment. Certain intensity of wind, frost, rain and snow, and small animal activities will trigger intrusion alarms, resulting in high false positives and false alarms. Single millimeter wave radar or video analysis technology has inherent defects and cannot fully meet the high-level security and protection requirements of the airport perimeter.

[0005] Therefore, in view of the defects in the prior art, a new type of airport perimeter intrusion alarm system needs to be developed. SUMMARY

[0006] In order to overcome the defects of the prior art, the application provides an airport perimeter intrusion alarm system based on millimeter wave radar and video analysis, which can not only detect intrusion behavior in real time and accurately, but also reduce false positives and false negatives, improve the automation and intelligence level of airport perimeter safety management, and provide solid protection for airport safety operation.

[0007] In order to achieve the above purpose, the application provides the following technical solutions:

[0008] An airport perimeter intrusion alarm system based on millimeter wave radar and video analysis, characterized in that it comprises:

[0009] a data acquisition module for scanning the airport perimeter area with a millimeter wave radar and receiving the echo signal of the target reflection to obtain target point cloud data, and for collecting video data of the airport perimeter area in real time with a camera;

[0010] a preprocessing and feature extraction module for denoising, filtering and clustering the target point cloud data, for denoising the video frames of the video data, and for inputting the processed target point cloud data and video frames into a ResNet structure network respectively to extract radar features and image features;

[0011] a multi-modal feature fusion module for using a Transformer network to construct the global correlation between radar features and image features through an attention mechanism, to dynamically and adaptively learn the relationship between radar features and image features, to realize the fusion of radar features and image features, and to obtain fusion features;

[0012] an intrusion target detection and recognition module for inputting the fusion features into a target detection head to identify and locate the intrusion target, and for using multiple classifiers to classify and determine the intrusion behavior;

[0013] an alarm and linkage module for alarming based on the detection and recognition results of the intrusion detection and recognition module, and for recording the relevant data of the intrusion behavior.

[0014] Preferably, the data acquisition module is also used for time synchronization and spatial calibration of the millimeter wave radar and the camera.

[0015] Preferably, the denoising, filtering and clustering of the target point cloud data specifically includes filtering out the noise points of the target point cloud data using a constant false alarm rate detection algorithm, and clustering combined with neighborhood density to remove isolated scattered points in the target point cloud data.

[0016] T = a P N ,

[0017]

[0018] wherein T is the detection threshold of the target point cloud data, P N is the noise power estimate value calculated based on the neighborhood reference unit, the coefficient a is used to ensure that the set false alarm probability is unchanged, N neighbor is the number of target point cloud data adjacent to the target point cloud data, and the points with a neighborhood radius of r and a point number less than p min are determined as noise.

[0019] Preferably, Gaussian filtering is adopted to perform noise reduction processing on the video frames of the video data.

[0020] Preferably, in the extraction of radar features and image features, a fusion residual attention module is used to strengthen the extraction of key features of the radar features and the image features.

[0021] Preferably, in the fusion of radar features and image features by using a Transformer network, a multi-level fusion structure of the Transformer network is adopted, and in each layer, intra-modal self-attention and cross-modal cross-attention are alternately performed.

[0022] Preferably, in the input of the fusion features into a target detection head to identify and locate the intrusion target, a query-based detection method is adopted, a fixed number of learnable queries Q are introduced, and the fusion features are interacted with the Transformer Decoder to output a plurality of candidate intrusion targets.

[0023] Preferably, the alarm is specifically sent when the prediction probability of the intrusion behavior is greater than a confidence threshold and the behavior classification result is clear, and the video data and the target point cloud data related to the intrusion behavior are stored and reported.

[0024] Preferably, the alarm information includes the time, location, and behavior of the intrusion.

[0025] In addition, the application also provides an airport perimeter intrusion alarm method based on millimeter wave radar and video analysis, characterized in that it comprises the following steps:

[0026] The millimeter wave radar scans the airport perimeter area and receives the echo signal reflected by the target to obtain the target point cloud data, and the camera collects the video data of the airport perimeter area in real time;

[0027] The target point cloud data is denoised, filtered and clustered, the video frames of the video data are denoised, and the processed target point cloud data and video frames are respectively input into a ResNet structure network to extract radar features and image features;

[0028] The Transformer network is used to dynamically and adaptively learn the relationship between the radar features and the image features by constructing the global correlation between the radar features and the image features through the attention mechanism, so as to realize the fusion of the radar features and the image features and obtain the fusion features;

[0029] The fusion features are input into a target detection head to identify and locate the intrusion target, and a multi-classifier is used to classify and determine the intrusion behavior;

[0030] Based on the detection and identification results of the intrusion detection and identification module, an alarm is given, and relevant data of the intrusion behavior is recorded.

[0031] Compared with the prior art, the millimeter wave radar and video analysis based airport perimeter intrusion alarm system has one or more of the following beneficial technical effects:

[0032] 1. The present application introduces millimeter wave radar and video analysis fusion technology to revolutionize the traditional airport perimeter intrusion alarm system, providing an all-weather, high-precision, intelligent security protection means. Millimeter wave radar has strong penetration, is not affected by light and weather conditions, and can continuously monitor the perimeter area in rain, fog, dust and other harsh environments; video analysis technology has the advantages of intuitive and high visualization, and can clearly present the characteristics and behavior of the intrusion target. The fusion of the two not only can detect intrusion behavior in real time and accurately, but also can reduce false positives and false negatives, improve the automation and intelligence level of airport perimeter safety management, and provide a solid guarantee for the safe operation of the airport.

[0033] 2. The present application uses millimeter wave radar as an active monitoring means, uses electromagnetic waves to detect target distance, speed and direction, and is almost not affected by bad weather such as cloud, rain, fog and haze, and can work all-weather. Millimeter wave radar also has the advantages of long detection distance, wide coverage, fast response speed, etc., which can detect intrusion objects at a certain distance outside the perimeter and realize early warning outside the perimeter; video monitoring can obtain intuitive images and detailed features of the intrusion target, support classification identification of personnel, animals, unmanned aerial vehicles and other targets, and facilitate security personnel to intuitively confirm the alarm. The fusion of the two kinds of data not only can eliminate the limitations of single data, but also can build an all-coverage, high-precision and low-false alarm airport perimeter alarm system. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a schematic diagram of the millimeter wave radar and video analysis based airport perimeter intrusion alarm system of the present application;

[0035] Figure 2 is a flowchart of the millimeter wave radar and video analysis based airport perimeter intrusion alarm method of the present application. DETAILED DESCRIPTION

[0036] Before any embodiments of the application are explained in detail, it is to be understood that the application is not limited in its application to the details of construction and the arrangement of components set forth in the following description or illustrated in the following drawings. The application is capable of other embodiments and of being practiced or of being carried out in various ways. Also, it is to be understood that the phraseology and terminology used herein is for the purpose of description and should not be regarded as limiting. The use of "including," "comprising," or "having" and variations thereof herein is meant to encompass the items listed thereafter and equivalents thereof as well as additional items. Unless specified or limited otherwise, the terms "mounted," "connected," "supported," and "coupled" and variations thereof are used broadly and encompass both direct and indirect mountings, connections, supports, and couplings. Further, "connected" and "coupled" are not restricted to physical or mechanical connections or couplings.

[0037] Also, in the disclosure of the application, the terms "longitudinal", "lateral", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, indicate the orientation or positional relationship shown in the drawings, which are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the above terms cannot be understood as limiting the application; secondly, the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of one element can be one, and in another embodiment, the number of the element can be multiple, the term "one" cannot be understood as limiting the number.

[0038] In view of the defects of the existing airport perimeter intrusion alarm system, the application proposes a comprehensive solution, which realizes high-precision airport perimeter intrusion alarm by fusing millimeter wave radar technology and video analysis technology, combining target detection, behavior recognition, data fusion algorithm, and alarm processing and linkage mechanism. First, the millimeter wave radar continuously scans the perimeter area, and real-time obtains the distance, speed, angle and other information of the target; at the same time, the video monitoring equipment collects image and video data of the perimeter area. Second, the point cloud and image data are respectively preprocessed such as denoising, and feature extraction is carried out based on ResNet network. Then, the attention mechanism is used for fusion of point cloud and image features, and the Transformer decoder is used to identify the intrusion target and analyze the intrusion behavior. Finally, if the intrusion behavior is detected, the alarm mechanism is triggered, and the specific intrusion data is recorded.

[0039] Figure 1 The constitution schematic diagram of the airport perimeter intrusion alarm system based on millimeter wave radar and video analysis of the application is shown. As shown in Figure 1 The airport perimeter intrusion alarm system based on millimeter wave radar and video analysis of the application includes:

[0040] 1. A data acquisition module.

[0041] The data acquisition module is configured to scan the airport perimeter area using a millimeter wave radar and receive echo signals reflected by targets to obtain target point cloud data, and configured to collect video data of the airport perimeter area in real time using a camera.

[0042] In the present application, the data acquisition module is configured to use frequency-modulated continuous wave millimeter wave radars and high-definition cameras, which are installed at equal intervals along the airport perimeter, to ensure that the millimeter wave radar beams cover the perimeter and the outside area without gaps. The millimeter wave radars output target point cloud data, while completing time synchronization and spatial calibration; the cameras have wide dynamic range and night vision capabilities, transmit video data, and process video frames to align with the millimeter wave radar timestamps.

[0043] Specifically, the present application is configured to use a plurality of frequency-modulated continuous wave millimeter wave radars and a plurality of high-definition cameras, which are installed at equal intervals on the fence posts or independent towers, at a height slightly higher than the top of the fence, so that the millimeter wave radar beams cover the perimeter and a certain area outside the perimeter without obstacles, and the adjacent detection fan areas overlap each other to avoid detection gaps.

[0044] The millimeter wave radars detect targets by emitting millimeter waves and receiving reflected signals, and output target point cloud data reflecting target distance, azimuth angle, radial velocity, and other information. Each millimeter wave radar covers an area of hundreds of meters, and can operate normally in complex conditions such as darkness and smoke, achieving all-weather monitoring of the perimeter area. After collecting the target point cloud data of the millimeter wave radars, time synchronization and spatial calibration functions need to be completed, the cameras are synchronized, and the conversion relationship between the millimeter wave radar coordinate system and the camera coordinate system is obtained through calibration P camera = RP radar + T, so as to facilitate subsequent data fusion, P camera and P radar represent the coordinates of the millimeter wave radar point in the camera coordinate system and the radar coordinate system, respectively, and R and T are rotation matrix and translation matrix.

[0045] To adapt to outdoor environments, the cameras need to have wide dynamic range and night vision capabilities (such as infrared or low-light imaging), and can provide clear video in strong light and at night. The data acquisition module needs to perform basic processing on the video frames of the video data, such as frame de-jittering and distortion correction, and align the video frames with the radar data according to the timestamps, to create conditions for multi-modal fusion.

[0046] 2. A pre-processing and feature extraction module.

[0047] The preprocessing and feature extraction module is used for denoising, filtering and clustering processing of the target point cloud data, denoising processing of the video frames of the video data, and inputting the processed target point cloud data and video frames into a ResNet structure network respectively to extract radar features and image features.

[0048] In the application, the preprocessing and feature extraction module processes different modal data characteristics, the target point cloud data is denoised and filtered by a constant false alarm rate detection algorithm, and clustered, the video frames are denoised by a Gaussian filter, both are input into a ResNet structure network to extract features, and a residual attention module is used to strengthen key feature extraction.

[0049] Specifically, the preprocessing aims to improve the quality of the original data and extract feature tensors or feature vectors that can be used for subsequent deep learning. The target point cloud data of the millimeter wave radar and the video data of the camera have different formats and noise characteristics, so they are respectively subjected to targeted preprocessing and feature extraction.

[0050] The target point cloud data is denoised and filtered, a constant false alarm rate detection algorithm is applied to filter out noise points, and clustering is performed in combination with neighborhood density to remove isolated scattered points, and the formula is expressed as follows:

[0051] T = a P N ,

[0052]

[0053] Wherein, T is the detection threshold of the target point cloud data, P N is the noise power estimation value calculated based on the neighborhood reference unit, the coefficient a is used to ensure that the set false alarm probability is unchanged, and by filtering out the target point cloud data exceeding T, the denoising and filtering of the target point cloud data can be realized. At the same time, the points with less than p min of the points in the neighborhood with a radius of r are determined as noise and are also removed.

[0054] Gaussian filter is applied to the video frames to reduce image noise:

[0055]

[0056] Wherein, (x, y) is the coordinate offset of the pixel point relative to the center of the filter; sigma is the standard deviation, which determines the width of the Gaussian distribution (the larger the sigma, the more significant the smoothing effect).

[0057] In feature extraction, bird's eye view (BEV) is used as a unified representation for the fusion of target point cloud data and video frames, which can maintain the geometric structure of millimeter wave radar and the semantic information of images. The processed target point cloud data and video frames are input into their respective backbones, which are the core structure of neural networks and are mainly used to extract the features of input data. Both backbones use ResNet structure, which is composed of multiple residual blocks to ensure effective transmission and learning of deep features. The basic unit can be represented as:

[0058]

[0059] where x is the input feature, is a residual function, and Wi represents the weight parameter in the i-th layer of convolution layer. By fusing the residual attention module, the focusing ability of the network on key features is further strengthened, and the expression quality and robustness of multi-modal information are improved.

[0060] 3. Multi-modal feature fusion module.

[0061] The multi-modal feature fusion module uses a Transformer network to construct the global association between radar features and image features through an attention mechanism, dynamically and adaptively learns the relationship between radar features and image features, and realizes the fusion of radar features and image features to obtain fusion features.

[0062] In the present application, the multi-modal feature fusion module uses a Transformer network architecture to construct the global association between modalities through an attention mechanism, dynamically and adaptively learns the relationship between different modal features, and realizes efficient and flexible information fusion.

[0063] Specifically, the Transformer network architecture is used to establish global association between two modal features (radar features and image features) through an attention mechanism, realizing information fusion. Compared with simple feature concatenation or early / late fusion scheme, Transformer can adaptively learn the association relationship between different modal features, improving the flexibility and effectiveness of fusion.

[0064] First, the radar BEV and image BEV features are aligned, and the radar and image are flattened into sequence form and learnable position encoding is added to preserve spatial position information:

[0065] X′ modality =X modality +P.

[0066] And the Transformer multi-level fusion structure is adopted, and in each layer, the intra-modal self-attention and the cross-modal cross-attention are alternately performed, and the local details and the global context are hierarchically fused.

[0067] 4. An intrusion target detection and recognition module.

[0068] The intrusion target detection and recognition module is used for inputting the fusion features into a target detection head to identify and locate the intrusion target, and using a multi-classifier to classify and determine the intrusion behavior.

[0069] In the application, the intrusion target detection and recognition module inputs the fusion features into the target detection head, adopts a query-based detection method to accurately locate the intrusion target, and uses a multi-classifier to classify and determine the target behavior.

[0070] Specifically, the fusion features are input into the target detection head to identify and locate the intrusion target, a query-based detection method is adopted, a fixed number of learnable queries Q are introduced, and the queries are interacted with the fused features through a Transformer decoder to output a plurality of candidate targets.

[0071] MultiHead(Q,K,V)=Concat(head1,head2,...,head h )W o ,

[0072]

[0073] After locating the intrusion target, the type of the intrusion behavior of the target can be further determined, and a multi-classifier is used to determine the behavior category of the detected target.

[0074] 5. An alarm and linkage module.

[0075] The alarm and linkage module is used for alarming based on the detection and recognition results of the intrusion detection and recognition module, and recording the related data of the intrusion behavior.

[0076] In the application, the alarm and linkage module is based on the results of the intrusion detection and recognition module, when the intrusion behavior is detected, an alarm device such as an audible and visual alarm is triggered to send an alarm signal to remind the security personnel, and the related data of the intrusion event is recorded, including time, position, target information, etc., which provides a basis for subsequent security analysis and event tracing.

[0077] Specifically, the alarm and linkage module comprehensively evaluates whether to trigger an alarm according to the output of the detection and recognition module. If the prediction probability of the target behavior judgment exceeds the confidence threshold and the behavior classification result is clear, it is determined that an intrusion behavior occurs, and an alarm information is sent to the security command center. The alarm content includes time, location (which area of the fence), target category and behavior, and the intrusion related video and point cloud data are stored and reported for further research and judgment by security personnel. In addition, for key areas, this module supports sound and light alarm. When receiving an intrusion event report, the nearby lighting equipment is started immediately, and the security personnel management system is connected to send alarm information to the handheld terminal of the security personnel closest to the intrusion location; the alarm event is also recorded and managed to generate an alarm log for subsequent query and analysis.

[0078] The airport fence intrusion alarm system based on millimeter wave radar and video analysis can construct an all-weather, high-precision and low false alarm rate intrusion fence alarm system by combining the data acquisition module, the preprocessing and feature extraction module, the multi-modal feature fusion module, the intrusion target detection and recognition module, and the alarm and linkage module. By preprocessing, calibrating and extracting features from millimeter wave radar data and video data, and using attention mechanism for feature level fusion, the fused features are output to multiple detection heads to detect whether there is an intrusion behavior. When an intrusion behavior is detected, the alarm mechanism is triggered, and the intrusion information is recorded.

[0079] Figure 2 The flowchart of the airport fence intrusion alarm method based on millimeter wave radar and video analysis of the application is shown. As shown in Figure 2 The airport fence intrusion alarm method based on millimeter wave radar and video analysis of the application includes the following steps:

[0080] 1. Data acquisition.

[0081] The millimeter wave radar scans the airport fence area and receives the echo signal reflected by the target to obtain the target point cloud data, and the camera real-time collects the video data of the airport fence area.

[0082] 2. Preprocessing and feature extraction.

[0083] The target point cloud data is denoised, filtered and clustered, the video frames of the video data are denoised, and the processed target point cloud data and video frames are respectively input into a ResNet structure network to extract radar features and image features.

[0084] 3. Multi-modal feature fusion.

[0085] The global correlation between the radar features and the image features is constructed by using a Transformer network through an attention mechanism, the relationship between the radar features and the image features is dynamically and adaptively learned, the fusion of the radar features and the image features is realized, and the fusion features are obtained.

[0086] 4. Intrusion target detection and identification.

[0087] The fusion features are input into a target detection head to identify and locate the intrusion target, and a multi-classifier is used to classify and determine the intrusion behavior.

[0088] 5. Alarm and linkage.

[0089] Based on the detection and identification results of the intrusion detection and identification module, alarm is performed, and relevant data of the intrusion behavior is recorded.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not a limitation on the protection scope of the present application. Based on the idea of the present application, those skilled in the art can modify or equivalently replace the technical solutions of the present application without departing from the essence and scope of the technical solutions of the present application.

Claims

1. An airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis, characterized in that, include: The data acquisition module is used to scan the airport perimeter area with millimeter-wave radar and receive the echo signals reflected by the target to obtain target point cloud data, and to collect video data of the airport perimeter area in real time using a camera. The preprocessing and feature extraction module is used to perform denoising, filtering and clustering on the target point cloud data, denoising on the video frames of the video data, and inputting the processed target point cloud data and video frames into a ResNet structure network to extract radar features and image features respectively. The multimodal feature fusion module uses a Transformer network to build a global association between radar features and image features through an attention mechanism, dynamically and adaptively learns the relationship between radar features and image features, and achieves the fusion of radar features and image features to obtain fused features. The intrusion target detection and recognition module is used to input fused features into the target detection head to identify and locate intrusion targets, and to use a multi-classifier to classify and determine intrusion behavior; The alarm and linkage module is used to issue an alarm based on the detection and identification results of the intrusion detection and identification module, and to record relevant data of the intrusion behavior.

2. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 1, characterized in that, The data acquisition module is also used to perform time synchronization and spatial calibration of the millimeter-wave radar and camera.

3. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 1, characterized in that, The denoising, filtering, and clustering processing of the target point cloud data specifically involves: using a constant false alarm rate (CFAR) detection algorithm to filter out noise points in the target point cloud data, and combining this with neighborhood density for clustering to remove isolated scattered points in the target point cloud data. T=α·P N , Where T is the detection threshold for the target point cloud data, and P N This is a noise power estimate calculated based on neighboring reference cells. The coefficient α is used to ensure that the set false alarm probability remains unchanged. N neighbor The number of target point cloud data adjacent to the target point cloud data is given by the number of points in a neighborhood of radius r that is less than ρ. min The point is identified as noise.

4. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 3, characterized in that, Gaussian filtering is used to reduce noise in the video frames of the video data.

5. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 3, characterized in that, When extracting radar and image features, a residual attention module is fused to enhance the extraction of key features from both radar and image features.

6. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 1, characterized in that, When using Transformer networks to fuse radar features and image features, a multi-level fusion structure of Transformer networks is adopted, with intra-modal self-attention and cross-modal cross-attention alternately performed in each layer.

7. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 1, characterized in that, When inputting fused features into the target detection head to identify and locate intrusion targets, a query-based detection method is adopted, which introduces a fixed number of learnable queries Q and interacts with the fused features through the Transformer Decoder to output several candidate intrusion targets.

8. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 1, characterized in that, The alarm is triggered as follows: if the predicted probability of the intrusion behavior exceeds the confidence threshold and the behavior classification result is clear, it is determined that an intrusion behavior has occurred, an alarm message is sent, and video data and target point cloud data related to the intrusion behavior are stored and reported at the same time.

9. The airport perimeter intrusion alarm system based on millimeter-wave radar and video analysis according to claim 8, characterized in that, The sent alarm information includes the time, location, type of target, and behavior of the intrusion.

10. An airport perimeter intrusion alarm method based on millimeter-wave radar and video analysis, characterized in that, Includes the following steps: Millimeter-wave radar is used to scan the airport perimeter area and receive the echo signals reflected by the target to obtain target point cloud data, and cameras are used to collect video data of the airport perimeter area in real time. The target point cloud data is subjected to denoising, filtering and clustering processing, the video frames of the video data are subjected to denoising processing, and the processed target point cloud data and video frames are respectively input into a ResNet structure network to extract radar features and image features. By using a Transformer network and an attention mechanism, a global association between radar features and image features is constructed. The relationship between radar features and image features is dynamically and adaptively learned to achieve the fusion of radar features and image features to obtain fused features. The fused features are input into the target detection head to identify and locate intrusion targets, and a multi-classifier is used to classify and determine the intrusion behavior; Based on the detection and identification results of the intrusion detection and identification module, an alarm is triggered and relevant data on the intrusion behavior is recorded.

Images