Matching method, device, aircraft and storage medium for aerial target tracking

Abstract

The present invention belongs to the field of target tracking technology, and discloses a matching method, device, aircraft and storage medium for aerial target tracking. The method includes: determining the tracking prediction state of the target at the current moment, and determining the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area; determining the center position similarity and size and posture similarity between the tracking prediction state and the measurement state; determining the cost function value according to the center position similarity and size and posture similarity; determining the matching result between the tracking prediction state and the measurement state based on the cost function value. In the above manner, the center position similarity and size and posture similarity between the tracking prediction state and the measurement state are comprehensively considered, the matching accuracy is improved, and the accuracy of target tracking is further improved.

Description

Technical Field

[0001] The present invention relates to the field of target tracking technology, and in particular to a matching method, device, aircraft and storage medium for aerial target tracking. Background Art

[0002] In existing target tracking solutions, the target matching cost function based on weighted Euclidean distance is generally used to match the measurement and trajectory. The specific method is to select feature vectors that can represent the target's position, speed, heading angle and other features, and calculate the weighted Euclidean distance between the measurement and trajectory. However, this method only considers the target's motion information and ignores the target's inherent properties such as size and shape, resulting in low matching accuracy, which affects the accuracy of target tracking.

[0003] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] The main purpose of the present invention is to provide a matching method, device, aircraft and storage medium for aerial target tracking, aiming to improve matching accuracy and improve target tracking accuracy.

[0005] To achieve the above object, the present invention provides a matching method for aerial target tracking, the method comprising the following steps:

[0006] Determine the tracking prediction state of the target at the current moment, and determine the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area;

[0007] Determine the center position similarity and size and posture similarity between the tracking prediction state and the measurement state;

[0008] Determine a cost function value according to the center position similarity and the size and posture similarity;

[0009] A matching result between the tracking prediction state and the measurement state is determined based on the cost function value.

[0010] Optionally, determining the center position similarity between the tracking prediction state and the measurement state includes:

[0011] Constructing a first Gaussian distribution obeyed by the prediction center position according to the tracking prediction state;

[0012] Constructing a second Gaussian distribution obeyed by the measurement center position according to the measurement state;

[0013] The center position similarity between the tracking prediction state and the measurement state is determined according to the first Gaussian distribution and the second Gaussian distribution.

[0014] Optionally, constructing a first Gaussian distribution obeyed by the prediction center position according to the tracking prediction state includes:

[0015] Determine a predicted three-dimensional bounding box according to the tracking prediction state;

[0016] Determine a first mean value according to the coordinates of the center point of the predicted three-dimensional bounding box;

[0017] Determine a first variance based on a one-step prediction value outputted from a preset filter according to the historical prediction value;

[0018] A first Gaussian distribution is constructed according to the first mean and the first variance.

[0019] Optionally, constructing a second Gaussian distribution obeyed by the measurement center position according to the measurement state includes:

[0020] Determine a measurement three-dimensional bounding box according to the measurement state;

[0021] Determine a second mean value according to the coordinates of the center point of the measured three-dimensional bounding box;

[0022] Determining, according to the measurement state, that the spatial position of the target object is in a detection area of ​​a corresponding sensor coordinate system;

[0023] Determine a second variance according to the detection characteristics of the sensor used in the detection area;

[0024] A second Gaussian distribution is constructed according to the second mean and the second variance.

[0025] Optionally, determining the center position similarity between the tracking prediction state and the measurement state according to the first Gaussian distribution and the second Gaussian distribution includes:

[0026] Determine a first overlapping area distribution according to the first Gaussian distribution and the second Gaussian distribution;

[0027] respectively calculating the rotation box volumes of the first Gaussian distribution, the second Gaussian distribution, and the first overlapping area distribution;

[0028] The overlap loss value is calculated according to the rotation box volumes respectively corresponding to the first Gaussian distribution, the second Gaussian distribution and the first overlapping area distribution, so as to obtain the center position similarity between the tracking prediction state and the measurement state.

[0029] Optionally, determining the size and posture similarity between the tracking prediction state and the measurement state includes:

[0030] Constructing a third Gaussian distribution obeyed by the predicted size and the predicted posture according to the tracking prediction state;

[0031] Constructing a fourth Gaussian distribution obeyed by the measurement size and the measurement posture according to the measurement state;

[0032] The size and posture similarity between the tracking prediction state and the measurement state is determined according to the third Gaussian distribution and the fourth Gaussian distribution.

[0033] Optionally, constructing a third Gaussian distribution obeyed by the predicted size and the predicted posture according to the tracking prediction state includes:

[0034] Determine a predicted three-dimensional bounding box according to the tracking prediction state;

[0035] Determine a third mean value according to the coordinates of the center point of the predicted three-dimensional bounding box;

[0036] Determining size information and posture information corresponding to the predicted three-dimensional bounding box;

[0037] Determine a predicted rotation matrix according to the posture information corresponding to the predicted three-dimensional bounding box, and determine a predicted diagonal matrix according to the size information corresponding to the predicted three-dimensional bounding box;

[0038] Determining a third difference based on the predicted rotation matrix and the predicted diagonal matrix;

[0039] A third Gaussian distribution is constructed according to the third mean and the third variance.

[0040] Optionally, constructing a fourth Gaussian distribution obeyed by the measurement size and the measurement posture according to the measurement state includes:

[0041] Determine a measurement three-dimensional bounding box according to the measurement state;

[0042] Determine a fourth mean value according to the coordinates of the center point of the measured three-dimensional bounding box;

[0043] Determine size information and posture information corresponding to the measured three-dimensional bounding box;

[0044] Determine a measurement rotation matrix according to the posture information corresponding to the measurement three-dimensional bounding box, and determine a measurement diagonal matrix according to the size information corresponding to the measurement three-dimensional bounding box;

[0045] determining a fourth variance based on the measurement rotation matrix and the measurement diagonal matrix;

[0046] A fourth Gaussian distribution is constructed according to the fourth mean and the fourth variance.

[0047] Optionally, the posture information includes at least a heading angle, a pitch angle, and a roll angle.

[0048] Optionally, determining the size and posture similarity between the tracking prediction state and the measurement state according to the third Gaussian distribution and the fourth Gaussian distribution includes:

[0049] Determine a second overlapping area distribution according to the third Gaussian distribution and the fourth Gaussian distribution;

[0050] respectively calculating the rotation box volumes of the third Gaussian distribution, the fourth Gaussian distribution, and the second overlapping area distribution;

[0051] The overlap loss value is calculated according to the rotation box volumes respectively corresponding to the third Gaussian distribution, the fourth Gaussian distribution and the second overlapping area distribution, so as to obtain the size and posture similarity between the tracking prediction state and the measurement state.

[0052] Optionally, determining the cost function value according to the center position similarity and the size and posture similarity includes:

[0053] A weighted average calculation is performed according to the center position similarity and the size and posture similarity to obtain a cost function value.

[0054] In addition, to achieve the above-mentioned purpose, the present invention also proposes a matching device for aerial target tracking, the matching device for aerial target tracking comprising:

[0055] A parameter determination module, used to determine the tracking prediction state of the target at the current moment, and determine the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area;

[0056] A similarity determination module, used to determine the center position similarity and size and posture similarity between the tracking prediction state and the measurement state;

[0057] A cost function determination module, used to determine a cost function value according to the center position similarity and the size and posture similarity;

[0058] A matching module is used to determine a matching result between the tracking prediction state and the measurement state based on the cost function value.

[0059] In addition, to achieve the above-mentioned objectives, the present invention also proposes an aircraft, which includes: a memory, a processor, and a matching program for aerial target tracking stored in the memory and executable on the processor, wherein the matching program for aerial target tracking is configured to implement the matching method for aerial target tracking as described above.

[0060] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which a matching program for aerial target tracking is stored. When the matching program for aerial target tracking is executed by a processor, the matching method for aerial target tracking as described above is implemented.

[0061] The present invention determines the tracking prediction state of the target at the current moment, and determines the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area; determines the center position similarity and size and posture similarity between the tracking prediction state and the measurement state; determines the cost function value according to the center position similarity and size and posture similarity; and determines the matching result between the tracking prediction state and the measurement state based on the cost function value. In the above manner, the center position similarity and size and posture similarity between the tracking prediction state and the measurement state are comprehensively considered, the matching accuracy is improved, and the high-precision matching result is used as the matching basis for target association in the aerial target tracking algorithm, further improving the accuracy of target tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic diagram of the structure of an aircraft in a hardware operating environment involved in an embodiment of the present invention;

[0063] Figure 2 It is a flow chart of a first embodiment of a matching method for aerial target tracking according to the present invention;

[0064] Figure 3 It is a flow chart of a second embodiment of a matching method for aerial target tracking according to the present invention;

[0065] Figure 4 A schematic diagram of the approximate calculation of the overlapping area distribution in the present invention;

[0066] Figure 5 It is a flow chart of a third embodiment of a matching method for aerial target tracking according to the present invention;

[0067] Figure 6 It is a schematic diagram of a specific process of calculating the cost function in the present invention;

[0068] Figure 7 It is a structural block diagram of the first embodiment of the matching device for aerial target tracking of the present invention.

[0069] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0070] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0071] Reference Figure 1 , Figure 1 A schematic diagram of the aircraft structure of the hardware operating environment involved in the embodiment of the present invention.

[0072] like Figure 1 As shown, the aircraft may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0073] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation of the aircraft, and may include more or less components than shown, or combine certain components, or arrange the components differently.

[0074] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a matching program for air target tracking.

[0075] exist Figure 1In the aircraft shown, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the aircraft of the present invention can be set in the aircraft, and the aircraft calls the matching program for aerial target tracking stored in the memory 1005 through the processor 1001, and executes the matching method for aerial target tracking provided in an embodiment of the present invention.

[0076] The embodiment of the present invention provides a matching method for aerial target tracking, referring to Figure 2 , Figure 2 The figure is a flow chart of the first embodiment of the matching method for aerial target tracking of the present invention.

[0077] In this embodiment, the matching method for aerial target tracking includes the following steps:

[0078] Step S10: determining the tracking prediction state of the target at the current moment, and determining the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area.

[0079] It can be understood that the executor of this embodiment is a matching device for aerial target tracking (hereinafter referred to as the matching device), which can be installed on aircraft such as drones and flying cars, and is specifically used to achieve aerial target tracking.

[0080] It should be noted that the matching device is provided with a tracking algorithm for predicting the motion state of the identified target. The aircraft is provided with a variety of sensors, such as cameras, radar probes, etc., through the sensor parameters collected by sensors such as cameras and radar probes, the target is identified and detected based on the sensor parameters, and the measurement state of the target is determined, including the detection characteristics of the target collected by the sensor and the used sensor in the corresponding detection area. Optionally, the matching device detects multiple aerial targets, and the target is any one of the multiple aerial targets. In a specific implementation, the matching device performs aerial target recognition and detection according to the sensor parameters collected at each moment, and adopts Kalman filtering or extended Kalman filtering or other filters based on Bayesian estimation theory, combined with the target recognition detection information output state estimation of multiple frames of historical moments, to obtain the tracking prediction state of the target at the current moment.

[0081] Step S20: Determine the center position similarity and size and posture similarity between the tracking prediction state and the measurement state.

[0082] It should be understood that the center position similarity is used to evaluate the similarity between the predicted target center point position and the measured target center point position. Optionally, the center position similarity between the tracking prediction state and the measured state is calculated using the Euclidean distance. Preferably, the center position similarity between the tracking prediction state and the measured state is calculated using the Gaussian distribution IOU loss function.

[0083] It should be noted that the size and posture similarity is used to evaluate the similarity between the predicted target size and posture and the measured target size and posture. Optionally, the Euclidean distance is used to calculate the size and posture similarity between the tracking prediction state and the measured state. Preferably, the Gaussian distribution IOU loss function is used to calculate the size and posture similarity between the tracking prediction state and the measured state.

[0084] Step S30: determining a cost function value according to the center position similarity and the size and posture similarity.

[0085] It can be understood that the cost function value is calculated by substituting the center position similarity and the size and posture similarity into the cost function. Optionally, the cost function value is obtained by taking the average value of the center position similarity and the size and posture similarity.

[0086] Specifically, the step S30 includes: performing weighted average calculation according to the center position similarity and the size and posture similarity to obtain a cost function value.

[0087] It should be noted that the cost function value is calculated by the following formula (1):

[0088]

[0089] Among them, L is the cost function value, L 1 is the center position similarity, L 2 is the size and posture similarity, K 1 is the weight corresponding to the center position similarity, K 2 is the weight corresponding to the size and posture similarity.

[0090] Step S40: determining a matching result between the tracking prediction state and the measurement state based on the cost function value.

[0091] In a specific implementation, the tracking algorithm set in this embodiment is to predict the possible state of the target object in the next frame, provide prior knowledge for the state estimation of the target object, and find the optimal target position in the current frame data collected by the sensor. Furthermore, the matching result is used to evaluate the state prediction confidence to correct the tracking prediction value of the target object state.

[0092] This embodiment determines the tracking prediction state of the target at the current moment, and determines the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area; determines the center position similarity and size and posture similarity between the tracking prediction state and the measurement state; determines the cost function value based on the center position similarity and size and posture similarity; determines the matching result between the tracking prediction state and the measurement state based on the cost function value. In the above manner, the center position similarity and size and posture similarity between the tracking prediction state and the measurement state are comprehensively considered, the matching accuracy is improved, and the high-precision matching result is used as the matching basis for target association in the aerial target tracking algorithm, further improving the accuracy of target tracking.

[0093] refer to Figure 3 , Figure 3 It is a flow chart of the second embodiment of the matching method for aerial target tracking of the present invention.

[0094] Based on the first embodiment, step S20 of the matching method for aerial target tracking in this embodiment includes:

[0095] Step S201: constructing a first Gaussian distribution obeyed by the prediction center position according to the tracking prediction state.

[0096] It should be understood that the tracking prediction state of the tracking trajectory obeys a certain Gaussian distribution. The parameters corresponding to the Gaussian distribution include mean and variance, where the mean represents the state estimation of the target and the variance represents the uncertainty of the state estimation.

[0097] Specifically, the step S201 includes: determining a predicted three-dimensional bounding box according to the tracking prediction state; determining a first mean according to the coordinates of the center point of the predicted three-dimensional bounding box; determining a first variance based on a one-step prediction value output by a preset filter according to historical prediction values; and constructing a first Gaussian distribution according to the first mean and the first variance.

[0098] It should be noted that the 3D bounding box of the target object is determined when it is in the tracking prediction state, and the center point of the 3D bounding box is selected as the mean vector of the Gaussian distribution, that is, μ = (x,,). The preset filter is the filter corresponding to the tracking algorithm, which can be a Kalman filter or an extended Kalman filter or other filters based on Bayesian estimation theory. The preset filter continuously uses the state of the previous moment to estimate the state of the next moment. Since the tracking algorithm uses the preset filter combined with multi-frame historical time series tracking information to give a state estimate with a small variance, for the target object corresponding to the tracking trajectory, the mean square error value reflecting the accuracy of the one-step prediction value is determined according to the one-step prediction value of the preset filter as the variance of the Gaussian distribution. In this embodiment, when calculating the similarity of the center position, the time series tracking information is added to improve the matching accuracy between the predicted value and the measured value.

[0099] Step S202: constructing a second Gaussian distribution obeyed by the measurement center position according to the measurement state.

[0100] Specifically, the step S202 includes: determining a measurement three-dimensional bounding box according to the measurement state; determining a second mean according to the coordinates of the center point of the measurement three-dimensional bounding box; determining that the spatial position of the target object is in the detection area of ​​the corresponding sensor coordinate system according to the measurement state; determining a second variance according to the detection characteristics of the sensor used in the detection area; and constructing a second Gaussian distribution according to the second mean and the second variance.

[0101] It should be understood that, similarly, the measurement state obtained by the sensor measurement obeys a certain Gaussian distribution. Determine the 3D bounding box when the target is in the measurement state, and select the center point of the 3D bounding box as the mean vector of the Gaussian distribution, that is, μ = (x, y, z). The sensor coordinate system is a coordinate system defined in advance to describe the position of the image point in the sensor space. For the measurement state obtained by the sensor measurement, the detection characteristics of the corresponding sensor need to be considered. The detection characteristics of sensors such as cameras and radar probes are related to the distance, orientation and angle corresponding to the target. In this embodiment, the detection area of ​​the sensor coordinate system is determined according to the detection characteristics of the sensor used in the detection area according to the detected spatial position of the target. The variance corresponding to the measurement state is determined, wherein the detection characteristics of the sensor are used to describe the corresponding relationship between the detection area and the variance value. In a specific implementation, the initialization variance corresponding to different detection areas in the sensor coordinate system is determined in advance through statistical experimental analysis and stored. After the distance, orientation and angle of the target relative to the aircraft are detected, it is converted to the sensor coordinate system, and the target is determined to be in the detection area of ​​the corresponding sensor coordinate system. The stored initialization variance is queried, and the queried value is used as the variance of the Gaussian distribution. In this embodiment, when calculating the center position similarity, time series tracking information is added to improve the matching accuracy between the predicted value and the measured value.

[0102] Step S203: determining the center position similarity between the tracking prediction state and the measurement state according to the first Gaussian distribution and the second Gaussian distribution.

[0103] It should be noted that, optionally, the KL divergence (relative entropy) is used to measure the matching degree between the first Gaussian distribution and the second Gaussian distribution. The greater the difference between the first Gaussian distribution and the second Gaussian distribution, the greater the KL divergence.

[0104] Furthermore, the step S203 includes: determining a first overlapping area distribution according to the first Gaussian distribution and the second Gaussian distribution; respectively calculating the rotation box volumes of the first Gaussian distribution, the second Gaussian distribution and the first overlapping area distribution; calculating the overlap loss value according to the rotation box volumes corresponding to the first Gaussian distribution, the second Gaussian distribution and the first overlapping area distribution, to obtain the center position similarity between the tracking prediction state and the measurement state.

[0105] It should be understood that referring to Figure 4 , Figure 4 This is a schematic diagram of the approximate calculation of the overlapping area distribution in the present invention, assuming that the first Gaussian distribution is represented by G A , the second Gaussian distribution is represented by G B In the specific implementation, the projection of the Gaussian distribution σ domain on the orthogonal hyperplanes of different coordinates can be regarded as an n-dimensional hypercube, that is, a rotation box. The volume of the rotation box is calculated by the following formula (2):

[0106]

[0107] Among them, V(G) represents the volume of the rotation box, G is used to distinguish different Gaussian distributions, and Σ represents the variance corresponding to the Gaussian distribution.

[0108] The traditional 3D-KFIoU method calculates the volume of the intersection of two 3D bounding boxes, and the 3D-KFIoU method is optimized in this embodiment: G A With G B The first overlapping area distribution between can be regarded as the linear minimum variance estimation of its distribution, avoiding complex solution calculations. Specifically, the first overlapping area distribution is determined by the following formulas (3)-(6):

[0109] G A∩B =αG A (μ A ,∑ A )G B (μ B ,∑ B )=G(μ ∑∩B ,∑ A∩B) (3);

[0110] μ A∩B =μ A +K(μ B -μ A ) (4);

[0111] ∑ A∩B =∑ A -K∑ A (5);

[0112] K=∑ A (∑ A +∑ B ) -1 (6);

[0113] Among them, μ A is the first Gaussian distribution G A The mean value of A is the first Gaussian distribution G A The variance of B is the second Gaussian distribution G B The mean value, ∑ B is the second Gaussian distribution G B The variance of .

[0114] The overlap loss function composed of the first Gaussian distribution and the second Gaussian distribution can be expressed as the following formula (7):

[0115]

[0116] Substitute the volume of the rotation box of the first Gaussian distribution, the volume of the rotation box of the second Gaussian distribution, and the volume of the rotation box of the first overlapping area distribution into formula (7), calculate the overlap loss value, and obtain the center position similarity L 1 .

[0117] Step S204: Determine the size and posture similarity between the tracking prediction state and the measurement state.

[0118] It should be noted that, optionally, the Euclidean distance is used to calculate the size and posture similarity between the tracking prediction state and the measurement state. Preferably, the Gaussian distribution IOU loss function is used to calculate the size and posture similarity between the tracking prediction state and the measurement state.

[0119] This embodiment constructs a first Gaussian distribution that the prediction center position obeys according to the tracking prediction state; constructs a second Gaussian distribution that the measurement center position obeys according to the measurement state; and determines the center position similarity between the tracking prediction state and the measurement state according to the first Gaussian distribution and the second Gaussian distribution. In the above manner, the center position similarity between the tracking prediction state and the measurement state is determined according to the Gaussian distribution that the center position obeys, thereby improving the accuracy of target matching.

[0120] refer to Figure 5 , Figure 5 It is a flow chart of the third embodiment of the matching method for aerial target tracking of the present invention.

[0121] Based on the first embodiment, step S20 of the matching method for aerial target tracking in this embodiment includes:

[0122] Step S205: Determine the center position similarity between the tracking prediction state and the measurement state.

[0123] Optionally, the Euclidean distance is used to calculate the center position similarity between the tracking prediction state and the measurement state. Preferably, the Gaussian distribution IOU loss function is used to calculate the center position similarity between the tracking prediction state and the measurement state.

[0124] Step S206: constructing a third Gaussian distribution obeyed by the predicted size and the predicted posture according to the tracking prediction state.

[0125] Specifically, the step S206 includes: determining a predicted three-dimensional bounding box according to the tracking prediction state; determining a third mean according to the coordinates of the center point of the predicted three-dimensional bounding box; determining size information and posture information corresponding to the predicted three-dimensional bounding box; determining a predicted rotation matrix according to the posture information corresponding to the predicted three-dimensional bounding box, and determining a predicted diagonal matrix according to the size information corresponding to the predicted three-dimensional bounding box; determining a third difference according to the predicted rotation matrix and the predicted diagonal matrix; and constructing a third Gaussian distribution according to the third mean and the third difference.

[0126] It should be understood that the 3D bounding box of the target object in the tracking prediction state is determined, and the 3D bounding box includes coordinate information, size information and attitude information. Among them, the size information includes length, width and height (l, w, h), and the attitude information includes at least the heading angle, pitch angle and roll angle (α, β, γ). The third mean is formed according to the coordinates of the center point of the predicted three-dimensional bounding box: μ = (x, y, z) T The predicted rotation matrix is ​​calculated by the following formula (8), the predicted diagonal matrix is ​​determined by the following formula (9), and the third difference is determined by the following formula (10):

[0127]

[0128]

[0129] ∑=RΛR T 10);

[0130] Among them, Σ represents the variance corresponding to the Gaussian distribution, R represents the rotation matrix, and R T represents a diagonal matrix.

[0131] The current matching method only considers the target's course angle rotation in the horizontal plane, while this embodiment expands the target's posture to a three-dimensional rotation space, using the three degrees of freedom of heading angle, pitch angle, and roll angle, which conforms to the motion characteristics of aerial targets and further improves the matching accuracy.

[0132] Step S207: constructing a fourth Gaussian distribution obeyed by the measurement size and the measurement posture according to the measurement state.

[0133] Specifically, the step S207 includes: determining a measurement three-dimensional bounding box according to the measurement state; determining a fourth mean according to the coordinates of the center point of the measurement three-dimensional bounding box; determining size information and posture information corresponding to the measurement three-dimensional bounding box; determining a measurement rotation matrix according to the posture information corresponding to the measurement three-dimensional bounding box, and determining a measurement diagonal matrix according to the size information corresponding to the measurement three-dimensional bounding box; determining a fourth variance according to the measurement rotation matrix and the measurement diagonal matrix; and constructing a fourth Gaussian distribution according to the fourth mean and the fourth variance.

[0134] It should be noted that the target object is detected by the sensor parameters to determine the 3D bounding box of the target object when it is in the measurement state. The 3D bounding box includes coordinate information, size information and attitude information. The size information includes length, width and height (l, w, h), and the attitude information includes at least heading angle, pitch angle and roll angle (α, β, γ). The fourth mean is formed according to the coordinates of the center point of the predicted three-dimensional bounding box: μ = (x, y, z) T The measurement rotation matrix is ​​calculated by the above formula (8), the measurement diagonal matrix is ​​determined by the above formula (9), and the fourth variance is determined by the above formula (10).

[0135] Step S208: determining the size and posture similarity between the tracking prediction state and the measurement state according to the third Gaussian distribution and the fourth Gaussian distribution.

[0136] It should be understood that, optionally, the KL divergence (relative entropy) is used to measure the degree of matching between the third Gaussian distribution and the fourth Gaussian distribution. The greater the difference between the third Gaussian distribution and the fourth Gaussian distribution, the greater the KL divergence.

[0137] Further, the step S208 includes: determining a second overlapping area distribution according to the third Gaussian distribution and the fourth Gaussian distribution; respectively calculating the rotation box volumes of the third Gaussian distribution, the fourth Gaussian distribution and the second overlapping area distribution; calculating the overlap loss value according to the rotation box volumes corresponding to the third Gaussian distribution, the fourth Gaussian distribution and the second overlapping area distribution, to obtain the size and posture similarity between the tracking prediction state and the measurement state.

[0138] It should be noted that, refer to Figure 4 , assuming that the third Gaussian distribution is represented by G A , the fourth Gaussian distribution is represented by G B , the volume of the rotating box is calculated by the above formula (2), the third Gaussian distribution G A With the fourth Gaussian distribution G B The second overlapping area distribution between can be regarded as the linear minimum variance estimation of its distribution. Specifically, the second overlapping area distribution is determined by the above formulas (3)-(6). Substitute the above formula (7) into the rotation box volume of the third Gaussian distribution, the rotation box volume of the fourth Gaussian distribution, and the rotation box volume of the second overlapping area distribution, calculate the overlap loss value, and obtain the center position similarity L 2 .

[0139] In one implementation, referring to Figure 6 , Figure 6 Schematic diagram of a specific process of calculating the cost function in the present invention; a prediction filter corresponding to the tracking algorithm is used to combine the target time series tracking information to calculate a one-step prediction value, determine the tracking prediction state of the target object, and determine the Gaussian distribution of the center position of the tracking target based on the one-step prediction value, and determine the Gaussian distribution of the size and posture of the tracking target based on the size information and posture information; modeling is performed based on sensor characteristics, and the Gaussian distribution of the center position of the measurement target is determined based on the area in the sensor coordinate system according to the parameters collected by the sensor, and target detection is performed based on the parameters collected by the sensor, and the size information and posture information of the measurement are determined based on the target detection, and the Gaussian distribution of the size and posture of the measurement target is determined; the overlap loss value is obtained based on the Gaussian distribution of the center position of the tracking target and the Gaussian distribution of the center position of the measurement target, and L is obtained. 1 According to the Gaussian distribution of the tracking target size and posture and the Gaussian distribution of the measurement target size and posture, the overlap loss value is obtained according to the optimized 3D-KFIoU method, and L is obtained. 2 ; for L 1 and L 2 ; Weighted average to obtain the final 3D matching cost function L.

[0140] In this embodiment, a third Gaussian distribution to which the predicted size and predicted posture are subject is constructed according to the tracking prediction state; a fourth Gaussian distribution to which the measured size and measured posture are subject is constructed according to the measurement state; and the size and posture similarity between the tracking prediction state and the measurement state is determined according to the third Gaussian distribution and the fourth Gaussian distribution. In the above manner, the size and posture similarity between the tracking prediction state and the measurement state is determined according to the Gaussian distribution to which the size and posture are subject, thereby improving the accuracy of target matching.

[0141] In addition, an embodiment of the present invention further proposes a storage medium, on which a matching program for aerial target tracking is stored. When the matching program for aerial target tracking is executed by a processor, the matching method for aerial target tracking as described above is implemented.

[0142] Since the storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought by the technical solutions of the above embodiments, which will not be described one by one here.

[0143] Reference Figure 7 , Figure 7 It is a structural block diagram of the first embodiment of the matching device for aerial target tracking of the present invention.

[0144] like Figure 7 As shown, the matching device for aerial target tracking proposed in the embodiment of the present invention includes:

[0145] The parameter determination module 10 is used to determine the tracking prediction state of the target at the current moment, and determine the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area.

[0146] The similarity determination module 20 is used to determine the center position similarity and size and posture similarity between the tracking prediction state and the measurement state.

[0147] The cost function determination module 30 is used to determine the cost function value according to the center position similarity and the size and posture similarity.

[0148] The matching module 40 is used to determine a matching result between the tracking prediction state and the measurement state based on the cost function value.

[0149] It should be understood that the above is only an example and does not constitute any limitation on the technical solution of the present invention. In specific applications, technicians in this field can make settings as needed, and the present invention does not limit this.

[0150] This embodiment determines the tracking prediction state of the target at the current moment, and determines the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area; determines the center position similarity and size and posture similarity between the tracking prediction state and the measurement state; determines the cost function value based on the center position similarity and size and posture similarity; determines the matching result between the tracking prediction state and the measurement state based on the cost function value. In the above manner, the center position similarity and size and posture similarity between the tracking prediction state and the measurement state are comprehensively considered, the matching accuracy is improved, and the high-precision matching result is used as the matching basis for target association in the aerial target tracking algorithm, further improving the accuracy of target tracking.

[0151] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0152] In addition, for technical details not described in detail in this embodiment, reference may be made to the matching method for aerial target tracking provided in any embodiment of the present invention, and will not be repeated here.

[0153] In one embodiment, the similarity determination module 20 is further used to construct a first Gaussian distribution obeyed by the predicted center position according to the tracking prediction state; construct a second Gaussian distribution obeyed by the measured center position according to the measurement state; and determine the center position similarity between the tracking prediction state and the measurement state according to the first Gaussian distribution and the second Gaussian distribution.

[0154] In one embodiment, the similarity determination module 20 is also used to determine a predicted three-dimensional bounding box based on the tracking prediction state; determine a first mean based on the coordinates of the center point of the predicted three-dimensional bounding box; determine a first variance based on a one-step prediction value output by a preset filter according to historical prediction values; and construct a first Gaussian distribution based on the first mean and the first variance.

[0155] In one embodiment, the similarity determination module 20 is further used to determine a measurement three-dimensional bounding box according to the measurement state; determine a second mean according to the coordinates of a center point of the measurement three-dimensional bounding box; determine that the spatial position of the target object is in a detection area of ​​a corresponding sensor coordinate system according to the measurement state; determine a second variance according to the detection characteristics of the sensor used in the detection area; and construct a second Gaussian distribution according to the second mean and the second variance.

[0156] In one embodiment, the similarity determination module 20 is also used to determine a first overlapping area distribution based on the first Gaussian distribution and the second Gaussian distribution; respectively calculate the rotation box volumes of the first Gaussian distribution, the second Gaussian distribution, and the first overlapping area distribution; calculate the overlap loss value based on the rotation box volumes corresponding to the first Gaussian distribution, the second Gaussian distribution, and the first overlapping area distribution, to obtain the center position similarity between the tracking prediction state and the measurement state.

[0157] In one embodiment, the similarity determination module 20 is further used to construct a third Gaussian distribution to which the predicted size and predicted posture obey according to the tracking prediction state; construct a fourth Gaussian distribution to which the measured size and measured posture obey according to the measurement state; and determine the size and posture similarity between the tracking prediction state and the measurement state according to the third Gaussian distribution and the fourth Gaussian distribution.

[0158] In one embodiment, the similarity determination module 20 is also used to determine a predicted three-dimensional bounding box based on the tracking prediction state; determine a third mean based on the coordinates of the center point of the predicted three-dimensional bounding box; determine size information and posture information corresponding to the predicted three-dimensional bounding box; determine a predicted rotation matrix based on the posture information corresponding to the predicted three-dimensional bounding box, and determine a predicted diagonal matrix based on the size information corresponding to the predicted three-dimensional bounding box; determine a third difference based on the predicted rotation matrix and the predicted diagonal matrix; and construct a third Gaussian distribution based on the third mean and the third difference.

[0159] In one embodiment, the similarity determination module 20 is also used to determine a measured three-dimensional bounding box according to the measurement state; determine a fourth mean according to the coordinates of the center point of the measured three-dimensional bounding box; determine size information and posture information corresponding to the measured three-dimensional bounding box; determine a measured rotation matrix according to the posture information corresponding to the measured three-dimensional bounding box, and determine a measured diagonal matrix according to the size information corresponding to the measured three-dimensional bounding box; determine a fourth variance according to the measured rotation matrix and the measured diagonal matrix; and construct a fourth Gaussian distribution according to the fourth mean and the fourth variance.

[0160] In one embodiment, the attitude information includes at least a heading angle, a pitch angle, and a roll angle.

[0161] In one embodiment, the similarity determination module 20 is also used to determine the second overlapping area distribution based on the third Gaussian distribution and the fourth Gaussian distribution; respectively calculate the rotation box volumes of the third Gaussian distribution, the fourth Gaussian distribution and the second overlapping area distribution; calculate the overlap loss value according to the rotation box volumes corresponding to the third Gaussian distribution, the fourth Gaussian distribution and the second overlapping area distribution, and obtain the size and posture similarity between the tracking prediction state and the measurement state.

[0162] In one embodiment, the cost function determination module 30 is further configured to perform weighted average calculation according to the center position similarity and the size and posture similarity to obtain a cost function value.

[0163] In addition, it should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0164] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0165] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0166] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A matching method for aerial target tracking, characterized in that: The matching method for aerial target tracking includes: Determine the tracking prediction state of the target at the current moment, and determine the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area; Determine the center position similarity and size and posture similarity between the tracking prediction state and the measurement state; Determine a cost function value according to the center position similarity and the size and posture similarity; Determine a matching result between the tracking prediction state and the measurement state based on the cost function value; Determining the center position similarity between the tracking prediction state and the measurement state includes: Determine a predicted three-dimensional bounding box according to the tracking prediction state; Determine a first mean value according to the coordinates of the center point of the predicted three-dimensional bounding box; Determine a first variance based on a one-step prediction value outputted from a preset filter according to the historical prediction value; constructing a first Gaussian distribution according to the first mean and the first variance; Constructing a second Gaussian distribution obeyed by the measurement center position according to the measurement state; The center position similarity between the tracking prediction state and the measurement state is determined according to the first Gaussian distribution and the second Gaussian distribution.

2. The matching method for aerial target tracking according to claim 1, characterized in that: The step of constructing a second Gaussian distribution obeyed by the measurement center position according to the measurement state includes: Determine a measurement three-dimensional bounding box according to the measurement state; Determine a second mean value according to the coordinates of the center point of the measured three-dimensional bounding box; Determining, according to the measurement state, that the spatial position of the target object is in a detection area of ​​a corresponding sensor coordinate system; Determine a second variance according to the detection characteristics of the sensor used in the detection area; A second Gaussian distribution is constructed according to the second mean and the second variance.

3. The matching method for aerial target tracking according to claim 1, characterized in that: The determining, according to the first Gaussian distribution and the second Gaussian distribution, the central position similarity between the tracking prediction state and the measurement state comprises: Determine a first overlapping area distribution according to the first Gaussian distribution and the second Gaussian distribution; respectively calculating the rotation box volumes of the first Gaussian distribution, the second Gaussian distribution, and the first overlapping area distribution; The overlap loss value is calculated according to the rotation box volumes respectively corresponding to the first Gaussian distribution, the second Gaussian distribution and the first overlapping area distribution, so as to obtain the center position similarity between the tracking prediction state and the measurement state.

4. The matching method for aerial target tracking according to claim 1, characterized in that: Determining the size and posture similarity between the tracking prediction state and the measurement state includes: Constructing a third Gaussian distribution obeyed by the predicted size and the predicted posture according to the tracking prediction state; Constructing a fourth Gaussian distribution obeyed by the measurement size and the measurement posture according to the measurement state; The size and posture similarity between the tracking prediction state and the measurement state is determined according to the third Gaussian distribution and the fourth Gaussian distribution.

5. The matching method for aerial target tracking according to claim 4, characterized in that: The step of constructing a third Gaussian distribution obeyed by the predicted size and the predicted posture according to the tracking predicted state includes: Determine a predicted three-dimensional bounding box according to the tracking prediction state; Determine a third mean value according to the coordinates of the center point of the predicted three-dimensional bounding box; Determining size information and posture information corresponding to the predicted three-dimensional bounding box; Determine a predicted rotation matrix according to the posture information corresponding to the predicted three-dimensional bounding box, and determine a predicted diagonal matrix according to the size information corresponding to the predicted three-dimensional bounding box; Determining a third difference based on the predicted rotation matrix and the predicted diagonal matrix; A third Gaussian distribution is constructed according to the third mean and the third variance.

6. The matching method for aerial target tracking according to claim 4, characterized in that: The step of constructing a fourth Gaussian distribution obeyed by the measurement size and the measurement posture according to the measurement state includes: Determine a measurement three-dimensional bounding box according to the measurement state; Determine a fourth mean value according to the coordinates of the center point of the measured three-dimensional bounding box; Determine size information and posture information corresponding to the measured three-dimensional bounding box; Determine a measurement rotation matrix according to the posture information corresponding to the measurement three-dimensional bounding box, and determine a measurement diagonal matrix according to the size information corresponding to the measurement three-dimensional bounding box; determining a fourth variance based on the measurement rotation matrix and the measurement diagonal matrix; A fourth Gaussian distribution is constructed according to the fourth mean and the fourth variance.

7. The matching method for aerial target tracking according to claim 5 or 6, characterized in that: The attitude information includes heading angle, pitch angle and roll angle.

8. The matching method for aerial target tracking according to claim 4, characterized in that: The determining the size and posture similarity between the tracking prediction state and the measurement state according to the third Gaussian distribution and the fourth Gaussian distribution includes: Determine a second overlapping area distribution according to the third Gaussian distribution and the fourth Gaussian distribution; respectively calculating the rotation box volumes of the third Gaussian distribution, the fourth Gaussian distribution, and the second overlapping area distribution; The overlap loss value is calculated according to the rotation box volumes respectively corresponding to the third Gaussian distribution, the fourth Gaussian distribution and the second overlapping area distribution, so as to obtain the size and posture similarity between the tracking prediction state and the measurement state.

9. The matching method for aerial target tracking according to any one of claims 1 to 6, characterized in that: The determining of the cost function value according to the center position similarity and the size and posture similarity includes: A weighted average calculation is performed according to the center position similarity and the size and posture similarity to obtain a cost function value.

10. A matching device for aerial target tracking, characterized in that: The matching device for aerial target tracking comprises: A parameter determination module, used to determine the tracking prediction state of the target at the current moment, and determine the corresponding measurement state of the target based on the spatial position of the target at the current moment and the detection characteristics of the sensor used in the corresponding detection area; A similarity determination module, used to determine the center position similarity and size and posture similarity between the tracking prediction state and the measurement state; A cost function determination module, used to determine a cost function value according to the center position similarity and the size and posture similarity; A matching module, used for determining a matching result between the tracking prediction state and the measurement state based on the cost function value; The similarity determination module is also used to determine a predicted three-dimensional bounding box based on the tracking prediction state; determine a first mean based on the coordinates of the center point of the predicted three-dimensional bounding box; determine a first variance based on a one-step prediction value output by a preset filter according to a historical prediction value; construct a first Gaussian distribution based on the first mean and the first variance; construct a second Gaussian distribution obeyed by the measurement center position based on the measurement state; and determine the center position similarity between the tracking prediction state and the measurement state based on the first Gaussian distribution and the second Gaussian distribution.

11. An aircraft, characterized in that: The aircraft includes: a memory, a processor, and a matching program for aerial target tracking stored in the memory and executable on the processor, wherein the matching program for aerial target tracking is configured to implement a matching method for aerial target tracking as described in any one of claims 1 to 9.

12. A storage medium, characterized in that: The storage medium stores a matching program for aerial target tracking, and when the matching program for aerial target tracking is executed by the processor, the matching method for aerial target tracking as described in any one of claims 1 to 9 is implemented.

Images