Vehicle-mounted unmanned aerial vehicle detection prevention and control system fusing multi-modal data

Abstract

The invention discloses a vehicle-mounted unmanned aerial vehicle detection prevention and control system fusing multi-modal data, and the system comprises a multi-modal parameter obtaining unit which is used for obtaining multi-modal behavior parameters observed by an unmanned aerial vehicle control terminal in a detection region; the initial time anchoring unit, the instantaneous anomaly degree construction unit, the instantaneous anomaly sequence construction unit and the double-sequence determination unit are used for determining a current anomaly sequence and a neighbor anomaly sequence on the instantaneous anomaly sequence; the dynamic threshold value construction unit is used for constructing a dynamic abnormal threshold value according to the current abnormal sequence and the neighbor abnormal sequence; a sudden change determination unit used for determining whether the unmanned aerial vehicle target has a significant behavior sudden change according to the dynamic abnormal threshold; according to the invention, by calculating the trend deviation included angle between the current abnormal sequence and the neighbor abnormal sequence, the behavior trend of the unmanned aerial vehicle can be noticed in time when the behavior trend is obviously changed, and the generated dynamic abnormal threshold improves the recognition capability of abnormal jump at the next moment.

Description

TECHNICAL FIELD

[0001] The present application relates to a UAV detection and prevention system, in particular to a vehicle-mounted UAV detection and prevention system fusing multi-modal data. BACKGROUND

[0002] In the existing vehicle-mounted UAV detection and prevention system, the multi-modal behavior parameters of the target are usually obtained through radar, radio frequency detection module and photoelectric imaging module, and based on these parameters, a fixed threshold or a dynamic threshold method based on historical data is set to evaluate the behavior risk of the UAV.

[0003] The prior art disclosed in patent publication No. CN119644323AXXX discloses a UAV detection method, device and system fusing multi-modal data, which fuses radar signals and sound wave signals to detect UAVs, improves the detection accuracy of UAVs, and accurately detects UAVs in overhead power transmission line areas.

[0004] However, such methods face significant problems in practical application:

[0005] On the one hand, when the UAV only slightly deviates from the normal flight mode (such as temporary hovering or small amplitude maneuvering), its parameters may exceed the fixed threshold and be misjudged as abnormal; and in emergency situations such as sudden acceleration of the target approaching or dramatic fluctuations in remote control signals, the fixed threshold cannot be dynamically adjusted with the evolution of behavior, resulting in delayed alarm.

[0006] On the other hand, if the dynamic threshold based on historical statistics is continuously used, although it can adapt to environmental changes to a certain extent, there is a risk of threshold drift. For example, when the UAV is in a non-threatening abnormal state for a long time (such as continuously hovering at low speed in the monitoring area), the system will include its behavior in statistical calculation, causing the moving average to gradually increase and the judgment threshold to move up, so that the subsequent real dangerous behavior mutation (such as suddenly rushing towards the sensitive area at high speed) may not be identified in time, causing a false negative.

[0007] More importantly, the existing method lacks the ability to sensitively identify changes in behavior trend evolution direction. When the UAV suddenly changes from slow approach to rapid approach, or from stable hovering to high-speed maneuvering, the evolution path of its behavior risk has obviously changed direction. However, relying only on the size of the instantaneous abnormal value at a certain moment, or analyzing the average trend of the abnormal degree over a period of time, it is difficult to accurately capture the "inflection point" in the trend. As a result, the system may not respond until a certain period of time after the behavior mutation occurs, seriously affecting the timeliness of the vehicle-mounted detection system. SUMMARY

[0008] In view of the deficiencies of the prior art, the application provides a vehicle-mounted unmanned aerial vehicle detection and prevention system fusing multi-modal data, trend deviation angle triggered dynamic abnormal threshold generation is achieved, two-stage linkage judgment from trend early warning to abnormal confirmation is achieved, so as to solve the technical problems proposed in the background art.

[0009] To achieve the above object, the application is implemented by the following technical solutions:

[0010] A vehicle-mounted unmanned aerial vehicle detection and prevention system fusing multi-modal data, the system comprises:

[0011] A multi-modal parameter acquisition unit is configured to acquire multi-modal behavior parameters observed by an unmanned aerial vehicle management terminal in a detection area.

[0012] An initial time anchoring unit is configured to anchor an initial timestamp of an unmanned aerial vehicle target based on the multi-modal behavior parameters.

[0013] A transient abnormality degree construction unit is configured to construct a transient abnormality degree of the unmanned aerial vehicle target at the initial timestamp according to the multi-modal behavior parameters of the unmanned aerial vehicle target at the initial timestamp.

[0014] A transient abnormality sequence construction unit is configured to continuously update transient abnormality degrees of the unmanned aerial vehicle target at N continuous timestamps starting from the initial timestamp, and construct a transient abnormality sequence.

[0015] A double-sequence determination unit is configured to determine a current abnormality sequence and a near-neighbor abnormality sequence on the transient abnormality sequence.

[0016] A dynamic threshold construction unit is configured to construct a dynamic abnormality threshold according to the current abnormality sequence and the near-neighbor abnormality sequence.

[0017] A mutation determination unit is configured to determine whether the unmanned aerial vehicle target has a significant behavior mutation according to the dynamic abnormality threshold.

[0018] In some specific embodiments, the application step of the initial time anchoring unit comprises:

[0019] S2-1, if any multi-modal behavior parameter exceeds its preset detection threshold, it is determined that there is an unmanned aerial vehicle target in the detection area.

[0020] S2-2, the multi-modal behavior parameters acquired when it is first determined that there is an unmanned aerial vehicle target are recorded as initial state parameters, and the corresponding timestamp is recorded as an initial timestamp.

[0021] In some specific embodiments, the application step of the transient abnormality degree construction unit comprises:

[0022] S3-1, extract the abnormal features of each modal parameter at the initial timestamp,

[0023] S3-2, define the instantaneous abnormality degree of the initial timestamp according to the abnormal feature of each modal parameter.

[0024] In some specific embodiments, the abnormal feature of each modal parameter at the initial timestamp is extracted, including:

[0025] S3-1-1, obtain the standard parameter threshold corresponding to the initial state parameter;

[0026] S3-1-2, calculate the deviation of the standard parameter threshold and the initial state parameter to obtain the parameter deviation of the initial timestamp;

[0027] S3-1-3, normalize the parameter deviation of the initial timestamp to extract the abnormal feature of each modal parameter at the initial timestamp;

[0028] In some specific embodiments, the instantaneous abnormality degree of the initial timestamp is defined according to the abnormal feature of each modal parameter, including:

[0029] S3-2-1, concatenate the abnormal features of each modal parameter in a predetermined order to form a multi-modal instantaneous vector;

[0030] S3-2-2, calculate the L2 norm of the multi-modal instantaneous vector, and define it as the instantaneous abnormality degree of the initial timestamp;

[0031] In some specific embodiments, the application step of the double sequence determination unit includes:

[0032] S5-1, in the instantaneous abnormality sequence, take the instantaneous abnormality degree corresponding to the latest timestamp as the endpoint, and select the instantaneous abnormality degrees corresponding to K consecutive timestamps in the direction of the initial timestamp to build a current time window with a length of K;

[0033] S5-2, build a neighbor time window with a length of K on the side of the current time window close to the initial timestamp and at a position Q timestamps away from it; Wherein, Q max ] and Q max <<K;

[0034] S5-3, after each update of the latest timestamp, calculate the K instantaneous abnormality degrees in the current time window and the neighbor time window, respectively;

[0035] S5-4, sort the K instantaneous abnormality degrees in the current time window and the neighbor time window according to the timestamps to form a current abnormality sequence and a neighbor abnormality sequence;

[0036] In some specific embodiments, the application step of the dynamic threshold construction unit includes:

[0037] S6-1, calculating a trend deviation angle according to the current anomaly sequence and the neighbor anomaly sequence;

[0038] S6-2, obtaining the dynamic anomaly threshold based on the trend deviation angle.

[0039] In some specific embodiments, calculating a trend deviation angle according to the current anomaly sequence and the neighbor anomaly sequence comprises:

[0040] S6-1-1, performing linear regression fitting on the current anomaly sequence and the neighbor anomaly sequence, extracting a trend change rate thereof to construct a current trend vector and a neighbor trend vector;

[0041] S6-1-2, calculating a trend deviation angle between the current trend vector and the neighbor trend vector;

[0042] In some specific embodiments, obtaining the dynamic anomaly threshold based on the trend deviation angle comprises:

[0043] S6-2-1, comparing the trend deviation angle with a predefined maximum allowed angle;

[0044] S6-2-2, if the trend deviation angle is greater than the maximum allowed angle, calculating a moving mean and a standard deviation of K instantaneous anomaly degrees in a current time window; otherwise, continuing to update the instantaneous anomaly degree;

[0045] S6-2-3, constructing a dynamic anomaly threshold for judging whether an instantaneous anomaly degree of a next timestamp is out of limit according to the moving mean and the standard deviation of the K instantaneous anomaly degrees in the current time window;

[0046] In some specific embodiments, the application step of the mutation determination unit comprises:

[0047] S7-1, obtaining an instantaneous anomaly degree corresponding to the next timestamp when the next timestamp arrives;

[0048] S7-2, comparing the instantaneous anomaly degree of the next timestamp with the dynamic anomaly threshold;

[0049] S7-3, if the instantaneous anomaly degree of the next timestamp is greater than the dynamic anomaly threshold, determining that the UAV target has a significant behavior mutation; otherwise, continuously updating the instantaneous anomaly sequence of the UAV target.

[0050] The present application provides a vehicle-mounted UAV detection and prevention system fusing multi-modal data, which has the following beneficial effects:

[0051] The present application can detect the change of the behavior trend of the UAV in time by calculating the trend deviation angle between the current abnormal sequence and the adjacent abnormal sequence. When the target suddenly changes from slowing down to fast approaching, or from stable flight to violent maneuvering, the direction of the trend change rate will change significantly, which is reflected in the increase of the angle between the trend vectors. Once the angle is detected to exceed the preset threshold, it is determined that the behavior trend has changed suddenly, and the high attention state is entered in advance.

[0052] Further, the present application only calculates the moving average and standard deviation based on the instantaneous abnormality in the current time window and constructs a dynamic abnormal threshold when the trend deviation angle exceeds the threshold. Since the threshold is not updated continuously, but triggered conditionally, the problem of gradually increasing threshold due to historical data accumulation when the target is in a non-threatening abnormal state for a long time (such as continuous circling) is avoided. The generated dynamic abnormal threshold can better reflect the current behavior characteristics of the target, and improve the recognition ability of the next moment abnormal jump. BRIEF DESCRIPTION OF DRAWINGS

[0053] Fig. 1 The structure block diagram of the vehicle-mounted UAV detection and prevention system of the present application fusing multi-modal data;

[0054] Fig. 2 The flowchart of the vehicle-mounted UAV detection and prevention system of the present application fusing multi-modal data;

[0055] Fig. 3 The construction flowchart of the current abnormal sequence and the adjacent abnormal sequence of the present application;

[0056] Fig. 4 The calculation flowchart of the trend deviation angle;

[0057] Fig. 5 The dynamic abnormal threshold diagram of the present application. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0059] Embodiment 1: Please refer to Figs. 1-2 The present application provides a vehicle-mounted UAV detection and prevention system fusing multi-modal data, which comprises the following units:

[0060] A multi-modal parameter acquisition unit is configured to acquire multi-modal behavior parameters observed by the UAV management and control terminal in the detection area;

[0061] Specifically, the multi-modal behavior parameters include an approaching speed measured by a radar, a remote control signal strength change rate captured by a radio frequency detection module, and a pixel size of the UAV in an optoelectronic imaging module, and the like. That is, the multi-modal behavior parameters cover motion trajectory features, electromagnetic behavior features, image features, and the like, and are typical multi-modal parameters.

[0062] An initial time anchoring unit is configured to anchor an initial timestamp of the UAV target based on the multi-modal behavior parameters;

[0063] A transient abnormality degree construction unit is configured to construct a transient abnormality degree of the UAV target at the initial timestamp according to the multi-modal behavior parameters of the UAV target at the initial timestamp;

[0064] A transient abnormality sequence construction unit is configured to construct a transient abnormality sequence by continuously updating transient abnormality degrees of the UAV target at N continuous timestamps starting from the initial timestamp. Specifically, the transient abnormality sequence reflects an abnormality level change trajectory of the UAV target after being detected for the first time.

[0065] A dual-sequence determination unit is configured to determine a current abnormality sequence and a neighbor abnormality sequence on the transient abnormality sequence;

[0066] A dynamic threshold construction unit is configured to construct a dynamic abnormality threshold according to the current abnormality sequence and the neighbor abnormality sequence;

[0067] A mutation determination unit is configured to determine whether the UAV target has a significant behavior mutation according to the dynamic abnormality threshold. In this embodiment, by acquiring multi-modal behavior parameters and anchoring an initial timestamp, continuously updating transient abnormality degrees to construct abnormality sequences, and then generating a dynamic abnormality threshold in combination with a current and neighbor abnormality sequence, determination of a significant behavior mutation of the UAV target is realized. This embodiment not only can reflect a process from appearance to behavior evolution of the UAV target, but also can dynamically adjust an abnormality threshold when a behavior trend deviates significantly, thereby improving the recognition ability of a sudden abnormal behavior, and enabling the vehicle-mounted detection system to have stronger adaptability in a complex environment.

[0068] Embodiment 2: refer to Figs. 3-5 The technical solution of this embodiment 2 is different from that of the embodiment 1 in that application steps of the units in the embodiment 1 are disclosed.

[0069] Exemplarily, the application steps of the initial time anchoring unit include:

[0070] S2-1, if any multi-modal behavior parameter exceeds a preset detection threshold, it is determined that the detection area has a UAV target;

[0071] That is, whether there is a UAV target depends on whether there is at least one parameter in the multi-modal behavior parameter exceeding its preset detection threshold. The preset detection threshold is calibrated according to the minimum detectable signal intensity of each sensor modality.

[0072] S2-2, record the multi-modal behavior parameters obtained when the UAV target is first determined to exist as initial state parameters, and record the corresponding time stamp as an initial time stamp;

[0073] Specifically, the initial time stamp represents the time when the UAV target is first detected and confirmed to exist, and is used as a time reference origin for multi-modal data evaluation.

[0074] Exemplarily, the application step of the instantaneous abnormality degree construction unit includes:

[0075] S3-1, extract the abnormal features of each modal parameter at the initial time stamp,

[0076] S3-2, define the instantaneous abnormality degree of the initial time stamp according to the abnormal features of each modal parameter.

[0077] Further, the step S3-1 specifically includes:

[0078] S3-1-1, obtain the standard parameter threshold corresponding to the initial state parameter;

[0079] Specifically, the standard parameter threshold represents the maximum typical value allowed for each multi-modal behavior parameter in a normal environment or non-threat state.

[0080] S3-1-2, calculate the deviation of the standard parameter threshold from the initial state parameter to obtain the parameter deviation of the initial time stamp;

[0081] Specifically, the parameter deviation is represented by the difference between the measured value of each multi-modal behavior parameter and its corresponding standard parameter threshold, and is used to quantify the deviation degree of the current observation state relative to the normal benchmark.

[0082] S3-1-3, normalize the parameter deviation of the initial time stamp to extract the abnormal features of each modal parameter at the initial time stamp;

[0083] Exemplarily, the normalization formula of the present embodiment is:

[0084]

[0085] wherein x i represents the i-th multi-modal parameter, x i，ref represents the standard parameter threshold corresponding to the i-th multi-modal parameter, and Δx i，maxmaximal allowed parameter deviation of the i-th multi-modal parameter, s i normalized anomaly feature of the i-th multi-modal parameter.

[0086] furthermore, x i -x i，ref parameter deviation between the i-th multi-modal parameter and the standard parameter threshold, reflecting the change of the current observed multi-modal parameter relative to the standard parameter threshold. max(0, ·) and min(1, ·) ensure that the final normalized anomaly feature is within the interval [0, 1]. If the ratio is less than 0, take 0; if greater than 1, take 1, which ensures that the anomaly features of all multi-modal behavior parameters fall on the same scale.

[0087] Exemplarily, assume that there is a radar measured speed of the approaching UAV as a multi-modal parameter, and its specific value is as follows:

[0088] the i-th multi-modal parameter x i = 7 m / s

[0089] the standard parameter threshold x i，ref = 5 m / s

[0090] the maximal allowed parameter deviation Δx i，max = 10 m / s

[0091] According to the formula:

[0092]

[0093] Therefore, based on the normalization formula, the normalized anomaly feature of each modal parameter can be obtained.

[0094] Further, the step S3-2 specifically comprises:

[0095] S3-2-1, splicing the anomaly feature of each modal parameter in a preset order to form a multi-modal instantaneous vector;

[0096] Specifically, the preset order can be a splicing rule preset based on the threat discrimination ability of each modal parameter, for example, placing a high confidence modal (such as radar speed measurement) at the front of the vector, or arranging according to the category order of motion features, electromagnetic features, and image features, to ensure that the multi-modal instantaneous vector constructed at different times has dimensional alignment.

[0097] S3-2-2, calculating the L2 norm of the multi-modal instantaneous vector, and defining it as the instantaneous anomaly degree of the initial timestamp;

[0098] The definition expression of the instantaneous anomaly is:

[0099]

[0100] wherein, represents the instantaneous abnormality degree of the initial timestamp, represents the multi-modal instantaneous vector of the initial timestamp, s i is the normalized abnormal feature of the i-th modal parameter, and n represents the number of modes.

[0101] Exemplarily, the L 2 norm represents the geometric length of the multi-modal instantaneous vector in the Euclidean space, reflecting the comprehensive parameter deviation of the abnormal features of each mode. Therefore, the L 2 norm can be used to characterize the overall abnormality degree of the multi-modal instantaneous vector. When multiple multi-modal parameters deviate from the normal state at the same time, even if a single deviation does not reach the alarm threshold, it can still be significantly amplified by the L 2 norm. Finally, the L 2 norm of the multi-modal instantaneous vector corresponding to the initial timestamp can be used as a comprehensive risk quantification index of the unmanned aerial vehicle target at that moment.

[0102] Exemplarily, the application steps of the double sequence determination unit include:

[0103] S5-1, in the instantaneous abnormality sequence, taking the instantaneous abnormality degree corresponding to the latest timestamp as the terminal point, selecting the instantaneous abnormality degrees corresponding to K consecutive timestamps in the direction of the initial timestamp along the time axis, and constructing a current time window with a length of K;

[0104] Specifically, the current time window focuses on the recent behavior evolution of the unmanned aerial vehicle target, and is updated in a sliding manner over time, which is used to capture potential threat acceleration or mode mutation.

[0105] S5-2, constructing a near neighbor time window with a length of K on the side of the current time window close to the initial timestamp and at a position away from the initial timestamp by Q timestamps; wherein Q∈[0, Q max ] and Q max <<K;

[0106] Specifically, the near neighbor time window represents a continuous time period located before the current time window and close in time, and has no overlap with the current time window, and can have a time gap or be directly adjacent, which is used to provide a recent behavior comparison. Further, the value of Q is used to control the time interval between the current time window and the near neighbor time window. When Q=0, it means that the near neighbor time window is close to the current time window, and there is no gap and no overlap between the two, forming a continuous time comparison interval.

[0107] S5-3, after each latest timestamp is updated, K instantaneous abnormality degrees in the current time window and the near neighbor time window are calculated respectively;

[0108] S5-4, sort the K instantaneous abnormality degrees in the current time window and the neighbor time window according to timestamps, to form a current abnormal sequence and a neighbor abnormal sequence;

[0109] Specifically, the sorting according to timestamps refers to arranging the instantaneous abnormality degrees in the window in order according to timestamps from early to late, to ensure that the sequence reflects the real behavior evolution trajectory; the current abnormal sequence corresponds to the current time window, and the neighbor abnormal sequence corresponds to the neighbor time window, both of which are ordered time sequences with a length of K.

[0110] Exemplarily, the application step of the dynamic threshold construction unit comprises:

[0111] S6-1, calculating a trend deviation angle according to the current abnormal sequence and the neighbor abnormal sequence;

[0112] S6-2, obtaining the dynamic abnormal threshold based on the trend deviation angle.

[0113] Further, the step S6-1 specifically comprises:

[0114] S6-1-1, performing linear regression fitting on the current abnormal sequence and the neighbor abnormal sequence, and extracting a trend change rate to construct a current trend vector and a neighbor trend vector;

[0115] Specifically, the step of linear regression fitting comprises:

[0116] For the abnormality degree sequence with a length of K, two variables are defined:

[0117] The input variable x represents the position index in the abnormality degree sequence, that is, X=[1, 2, …, K]. Each value in the position index represents the continuous timestamp at the corresponding position in the abnormality degree sequence.

[0118] The output variable Y represents the instantaneous abnormality degree corresponding to the continuous timestamp.

[0119] Then a linear model Y=aX+b is fitted by the least square method; wherein:

[0120] The slope a represents the trend change rate in the sequence, reflecting the change trend of the sequence degree with time, and the intercept b is the offset value when x=0.

[0121] Further, the least square method in this embodiment is to find a straight line that best represents the change trend of the abnormality degree.

[0122] Specifically:

[0123] The input is the time point (1st, 2nd, …, Kth), and the output is the instantaneous abnormality degree at the corresponding time. The least square method adjusts the slope and intercept of the straight line so that the sum of the square of the vertical deviation between the straight line and all data points is minimized. In this embodiment, the slope of the straight line is the trend change rate of the instantaneous abnormality degree in the time window.

[0124] Slope is positive: instantaneous abnormality degree shows an upward trend, and the target behavior of the UAV may tend to be dangerous;

[0125] Slope is negative: instantaneous abnormality degree is decreasing, and the target of the UAV may be away from or weaken the threat;

[0126] The greater the slope, the more dramatic the change.

[0127] It should be noted that although the trend change rate is a scalar (single value), in this embodiment, it is abstracted as a trend vector with directional characteristics by constructing a one-dimensional vector with it as the only component, so that the upward or downward trend of the trend can be formalized as the positive or negative direction in the vector space.

[0128] Current trend change rate = 0.8 → current trend vector = [0.8];

[0129] Neighbor trend change rate = 0.2 → neighbor trend vector = [0.2].

[0130] S6-1-2, calculate the trend deviation angle between the current trend vector and the neighbor trend vector;

[0131] The calculation formula of the trend deviation angle is:

[0132]

[0133] wherein, V curr represents the current trend vector, which is the vectorization representation of the trend change rate in the current abnormal sequence, v hist represents the neighbor trend vector, which is the vectorization representation of the trend change rate in the neighbor abnormal sequence, ||v curr || represents the L2 norm of the current trend vector, ||v hist || represents the L2 norm of the neighbor trend vector, and θ represents the trend deviation angle between the current trend vector and the neighbor trend vector.

[0134] Further, the trend deviation angle reflects the consistency of the current behavior trend and the neighbor trend in the evolution direction.

[0135] Specifically, the current trend vector and the neighbor trend vector are the vectorized representations of the trend change rates of the current anomaly sequence and the neighbor anomaly sequence, respectively, and the direction is determined by the positive and negative of the change rate - a positive value indicates that the abnormality degree is rising (e.g., the target is accelerating to approach), and a negative value indicates that the abnormality degree is falling (e.g., the target is decelerating or moving away).

[0136] The geometric meaning of the vector angle is the degree of deviation between the two directions:

[0137] When the two vectors are in the same direction (the angle is close to 0°), it means that the target risk trend continues to strengthen or weaken, and the behavior is continuous.

[0138] When the two vectors are in opposite directions (the angle is close to 180°), it means that the target changes from "stabilization" to "approach" or from "approach" to "withdrawal", and the behavior has a directional mutation.

[0139] The larger the angle, the more significant the trend deviation.

[0140] Therefore, the trend deviation angle converts the trend change of the time series into a measurable relationship through the direction comparison in the vector space, thereby effectively representing the mutation degree of the UAV target behavior.

[0141] Further, the step S6-2 specifically comprises:

[0142] S6-2-1, comparing the trend deviation angle with a predefined maximum allowed angle;

[0143] S6-2-2, if the trend deviation angle is greater than the maximum allowed angle, calculating the moving mean and standard deviation of the K instantaneous abnormality degrees in the current time window; otherwise, continue to update the instantaneous abnormality degree.

[0144] Specifically, the moving mean can reflect the overall abnormality level of the UAV target behavior in the current time period, and is used to represent the average abnormality strength; and the standard deviation can reflect the fluctuation degree of the instantaneous abnormality degree in the time period, and represents the instability of the target behavior.

[0145] S6-2-3, constructing a dynamic abnormality threshold for judging whether the next time stamp instantaneous abnormality degree is out of limit according to the moving mean and standard deviation of the K instantaneous abnormality degrees in the current time window;

[0146] Specifically, the dynamic abnormality threshold calculation logic is: taking the moving mean as a static sub-item, and adding a dynamic sub-item proportional to the standard deviation, and the proportional coefficient of the dynamic sub-item is a preset sensitivity coefficient.

[0147] Exemplarily, the moving average of the K instantaneous abnormal degrees in the current time window is taken as a static sub-item, representing the benchmark level of the target behavior risk; the standard deviation is taken as the calculation basis of a dynamic sub-item, reflecting the fluctuation characteristics of the abnormal degree sequence; the standard deviation is weighted by a preset sensitivity coefficient to obtain a dynamic sub-item, and the dynamic sub-item is added to the static sub-item to construct a dynamic abnormal threshold. The threshold is dynamically adjusted according to the multi-modal parameters of the unmanned aerial vehicle target, and is used to judge whether the instantaneous abnormal degree of the next time stamp is out of limit.

[0148] Exemplarily, the application step of the mutation determination unit comprises:

[0149] S7-1, obtaining the instantaneous abnormal degree corresponding to the next time stamp when the next time stamp arrives;

[0150] S7-2, comparing the instantaneous abnormal degree of the next time stamp with the dynamic abnormal threshold;

[0151] S7-3, if the instantaneous abnormal degree of the next time stamp is greater than the dynamic abnormal threshold, it is determined that the unmanned aerial vehicle target has a significant behavior mutation, otherwise, the instantaneous abnormal sequence of the unmanned aerial vehicle target is continuously updated;

[0152] It should be noted that the dynamic abnormal threshold is not compared with the instantaneous abnormal degree in the current time window or the adjacent time window, but compared with the instantaneous abnormal degree newly generated at the next time stamp. The purpose is:

[0153] When the behavior trend is detected to possibly have a mutation, a dynamic determination boundary is constructed based on the moving average and the standard deviation of the current time window, for evaluating whether the behavior at the next time instant exceeds the current expected range.

[0154] The detection and prevention system of the application first identifies whether the behavior evolution direction has a significant change through the trend deviation angle, and then combines the current behavior statistics to judge whether the instantaneous abnormal degree exceeds the expected range. The step-by-step processing mode enables the system to capture the turning point in the trend and verify whether there is an actual risk jump, effectively distinguishing between ordinary behavior fluctuations and threatening mutation behaviors, and reducing false positives and false negatives.

[0155] In the vehicle-mounted mobile detection scene, in the face of many environmental disturbances and uncertain target behaviors, the system does not rely on fixed models or artificially set rules, but autonomously decides whether to start the dynamic criterion according to the actually observed abnormal degree change, and can maintain stable recognition ability for significant behavior mutations of the unmanned aerial vehicle under complex conditions.

[0156] The above-described embodiments can be implemented in part or in whole through software, hardware, firmware or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product that includes one or more computer instructions or computer programs. When loaded and executed by a computer, the computer instructions or computer programs can produce the processes or functions described above in accordance with the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, from a website, a computer, a server or a data center to another website, computer, server or data center through a wired (for example, infrared, wireless, microwave or the like) manner.

[0157] The computer-readable storage medium can be any available medium or a collection of medium accessible by a computer or a data storage device such as a server, a data center or the like that includes one or more of the available medium. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0158] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described device embodiments are merely illustrative, for example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0159] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A vehicle-mounted drone detection and control system integrating multimodal data, characterized in that: include: A multimodal parameter acquisition unit is used to obtain multimodal behavior parameters observed by the drone control terminal in the detection area; The initial time anchoring unit is used to anchor the initial timestamp of the UAV target based on the multimodal behavior parameters; An instantaneous abnormality degree construction unit is used to construct the instantaneous abnormality degree of the UAV target at the initial timestamp based on the multimodal behavior parameters of the UAV target at the initial timestamp; The instantaneous anomaly sequence construction unit is used to continuously update the instantaneous anomaly degree of the UAV target at N consecutive timestamps with the initial timestamp as the starting point to construct the instantaneous anomaly sequence; A dual sequence determination unit, configured to determine a current abnormal sequence and a neighboring abnormal sequence based on an instantaneous abnormal sequence; A dynamic threshold construction unit is used to construct a dynamic anomaly threshold based on the current anomaly sequence and the neighboring anomaly sequence; The mutation determination unit is used to determine whether a significant behavioral mutation occurs in the UAV target based on the dynamic abnormality threshold.

2. The vehicle-mounted drone detection and control system integrating multimodal data according to claim 1 is characterized in that: The application steps of the initial time anchor unit include: S2-1. If any multimodal behavior parameter exceeds its preset detection threshold, it is determined that there is a drone target in the detection area; S2-2. The multimodal behavior parameters obtained when the presence of the drone target is first determined are recorded as initial state parameters, and the corresponding timestamp is recorded as the initial timestamp.

3. The vehicle-mounted drone detection and control system integrating multimodal data according to claim 2 is characterized in that: The application steps of the instantaneous abnormality construction unit include: S3-1, extract the abnormal features of each modal parameter at the initial timestamp, S3-2. Based on the abnormal characteristics of each modal parameter, define the instantaneous abnormality degree of the initial timestamp.

4. The vehicle-mounted drone detection and control system integrating multimodal data according to claim 3 is characterized in that: Extract abnormal features of each modal parameter at the initial timestamp, including: S3-1-1. Obtain the standard parameter threshold corresponding to the initial state parameter; S3-1-2. Calculate the deviation between the standard parameter threshold and the initial state parameter to obtain the parameter deviation of the initial timestamp; S3-1-3. Normalize the parameter deviation of the initial timestamp and extract the abnormal features of each modal parameter at the initial timestamp.

5. The vehicle-mounted UAV detection and control system integrating multimodal data according to claim 4 is characterized in that: According to the abnormal characteristics of each modal parameter, the instantaneous abnormality of the initial timestamp is defined, including: S3-2-1. Splice the abnormal features of each modal parameter in a preset order to form a multi-modal instantaneous vector; S3-2-2. Calculate the L2 norm of the multimodal instantaneous vector and define it as the instantaneous abnormality of the initial timestamp.

6. The vehicle-mounted UAV detection and control system integrating multimodal data according to claim 5 is characterized in that: The application step of the dual sequence determination unit includes: S5-1. In the instantaneous anomaly sequence, take the instantaneous anomaly degree corresponding to the latest timestamp as the end point, select the instantaneous anomaly degrees corresponding to K consecutive timestamps along the pseudo-time axis in the direction of the initial timestamp, and construct a current time window of length K; S5-2. Construct a neighboring time window of length K on the side of the current time window close to the initial timestamp and at a distance of Q timestamps from it; where Q∈[0,Q max ], and Q max <<K; S5-3. After each latest timestamp update, calculate the K instantaneous abnormality degrees in the current time window and the neighboring time windows respectively; S5-4. Sort the K instantaneous abnormality degrees in the current time window and the neighboring time window in order of timestamps to form a current abnormality sequence and a neighboring abnormality sequence.

7. The vehicle-mounted drone detection and control system integrating multimodal data according to claim 1 is characterized in that: The application steps of the dynamic threshold construction unit include: S6-1. Calculate the trend deviation angle based on the current abnormal sequence and the neighboring abnormal sequence; S6-2. Obtain the dynamic anomaly threshold based on the trend deviation angle.

8. The vehicle-mounted UAV detection and control system integrating multimodal data according to claim 7 is characterized in that: Calculate the trend deviation angle based on the current abnormal sequence and the neighboring abnormal sequence, including: S6-1-1. Perform linear regression fitting on the current abnormal sequence and the nearest abnormal sequence, extract their trend change rates, and construct the current trend vector and the nearest trend vector; S6-1-2. Calculate the trend deviation angle between the current trend vector and the nearest neighbor trend vector.

9. The vehicle-mounted UAV detection and control system integrating multimodal data according to claim 8 is characterized in that: The dynamic anomaly threshold is obtained based on the trend deviation angle, including: S6-2-1. Compare the trend deviation angle with a predefined maximum allowable angle; S6-2-2. If the trend deviation angle is greater than the maximum allowable angle, calculate the moving mean and standard deviation of K instantaneous anomalies in the current time window; otherwise, continue to update the instantaneous anomaly; S6-2-3. Based on the moving mean and standard deviation of K instantaneous anomalies in the current time window, a dynamic anomaly threshold is constructed to determine whether the instantaneous anomaly at the next timestamp exceeds the limit.

10. According to the vehicle-mounted UAV detection and control system integrating multimodal data as claimed in claim 9, the application steps of the mutation determination unit include: S7-1. Obtain the instantaneous abnormality corresponding to the arrival of the next timestamp; S7-2, comparing the instantaneous abnormality degree of the next timestamp with the dynamic abnormality threshold; S7-3. If the instantaneous abnormality of the next timestamp is greater than the dynamic abnormality threshold, it is determined that the drone target has a significant behavioral mutation; otherwise, the instantaneous abnormality sequence of the drone target is continuously updated.

Images