A partial occlusion extended target tracking method and device based on a random matrix method

Abstract

The application discloses a kind of based on random matrix method's partially occluded extended target tracking method and device, comprising: receiving the measurement data of initial time, initialize target state and extended form;With random matrix method, the state and extended form of target track of last time are predicted in one step;With radar position and obstacle position, shape judge whether there is occlusion;Receive the measurement data of current time, use random matrix method to update the state and extended form of unoccluded area, update unoccluded scale factor;With the state and extended form of unoccluded scale factor and unoccluded area, update overall target;Output target track information.The application improves the precision of state, extended form estimation under the condition that target is occluded, reduces the estimation error of target state, can be applied in intelligent transportation, crowded road target tracking in the field such as automatic driving, etc., accurately track occluded target.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of extended target tracking, and particularly relates to an extended target tracking method and device for the case that an extended target is partially occluded by an obstacle and no radar measurement is available in the occluded area. BACKGROUND

[0002] Target tracking is a very important supporting technology, and is crucial to information processing systems for advanced complex applications, and is widely used in missile guidance and anti-missile systems, automatic driving systems and other military and civilian fields, and has important application value.

[0003] Sensors for target tracking, including radars, sonars, optical instruments, etc., can provide measurement data with uncertain interference about the number, position, speed and other characteristics of the target. Compared with other sensors, millimeter wave radars can work all day and all night, have stable performance, are less affected by weather, are anti-jamming and anti-stealth, and are low in price, and are the first choice for detecting medium and long distance targets in the field of target tracking.

[0004] In real life, target tracking will encounter a variety of special situations, and occlusion is a common special situation. The target to be tracked may be occluded by fixed obstacles, other targets, etc., resulting in a large error in the estimation of the target position and extended shape when directly using existing algorithms, so that the occlusion of the target may cause serious consequences such as vehicle collision, endangering life safety.

[0005] Millimeter wave radar has few echoes, dense clutter and no imaging capability, and the partial occlusion target tracking based on millimeter wave radar faces the following difficulties: (1) multiple measurements of the target are generated, how to use these measurements to estimate the state and extended shape of the target; (2) the occluded area and proportion of the target are unknown; (3) how to use the partial measurements of the target to estimate the state and extended shape of the whole target.

[0006] The above difficulties can be summarized as the problem of extended target tracking under the condition of partial occlusion, that is, the state and extended shape of the target need to be estimated in real time, it is judged whether the target is occluded by an obstacle, the occluded area and proportion of the target are estimated, and the state and extended shape of the whole target are estimated by using the partial measurements of the target. However, the traditional target tracking theory and technology are usually for point targets, and the existing methods for extended target tracking do not consider the problem of target occlusion, and cannot be directly used to solve the problem of extended target tracking under the condition of partial occlusion. Therefore, on the basis of the traditional extended target tracking technology, a new extended target tracking algorithm suitable for the case that the target is partially occluded is proposed.

[0007] Therefore, it is a technical problem to be solved to provide an extended target tracking method for the case that the target is partially occluded. SUMMARY

[0008] The present application aims to solve the problem that the existing tracking algorithm cannot realize the extended target tracking in the case of occlusion. Based on the existing extended target tracking method based on random matrix, the present application innovatively introduces occlusion modeling, target un-occlusion proportion factor estimation and other means, realizes the use of partial measurement of the target by radar, i.e. the measurement of the un-occluded area of the target by radar, realizes the estimation of the overall target state and the extended shape, and the estimation of the proportion of the target being occluded. The present application finally realizes the accurate tracking of the extended target in the case of partial occlusion.

[0009] The present application is realized by the following technical solutions.

[0010] In one aspect of the present application, a partial occlusion extended target tracking method based on the random matrix method is provided, comprising:

[0011] Obtaining the measurement data at the initial time, initializing the target state and the extended shape;

[0012] Using the random matrix method to perform one-step prediction of the state and the extended shape of the target track at the previous time;

[0013] Using the radar position and the obstacle position and shape to determine whether there is occlusion;

[0014] If there is occlusion, calculating the un-occlusion proportion factor, and performing one-step prediction of the state and the extended shape of the un-occluded area;

[0015] Receiving the measurement data at the current time, using the random matrix method to update the state and the extended shape of the un-occluded area, and updating the un-occlusion proportion factor;

[0016] Using the updated un-occlusion proportion factor and the state and the extended shape of the un-occluded area to update the overall target;

[0017] Outputting the target track.

[0018] In the embodiments of the present application, receiving the measurement data at the initial time, initializing the target state and the extended shape, comprises:

[0019] Converting the sensor measurement in the polar coordinate system at the initial time to the rectangular coordinate system;

[0020] Using the mean and dispersion matrix of the measurement to initialize the target state and the extended shape.

[0021] In the embodiments of the present application, using the random matrix method to perform one-step prediction of the state and the extended shape of the target track at the previous time, comprises:

[0022] Obtain a motion model of the target, and perform one-step prediction on a state of a target track at a previous moment by using a random matrix method and the motion model;

[0023] Perform one-step prediction on an extended shape of the target track at the previous moment by using the random matrix method, describe and depict the extended shape changing over time by using an extended shape evolution model, and perform one-step prediction on the extended shape.

[0024] In the embodiment of the present application, whether there is an occlusion is determined by using the radar position and the obstacle position and shape, including:

[0025] The occluded area range is calculated by using the radar position and the obstacle position;

[0026] Whether there is an occlusion is determined according to the occluded area range and one-step prediction values of the target state and the extended shape.

[0027] In the embodiment of the present application, whether there is an occlusion is determined according to the occluded area range and one-step prediction values of the target state and the extended shape, including:

[0028] Four vertices of the target are calculated according to the target state and the one-step prediction values of the extended shape;

[0029] According to the vertex set of the target and the radar position and the radar pointing angle, the counterclockwise maximum angle and the clockwise maximum angle of the target relative to the radar pointing angle are obtained;

[0030] Whether there is an occlusion is determined according to the obstacle counterclockwise maximum angle and the obstacle clockwise maximum angle, the target counterclockwise maximum angle relative to the radar pointing angle, and the target clockwise maximum angle relative to the radar pointing angle.

[0031] In the embodiment of the present application, the unoccluded proportion factor is calculated, and one-step prediction is performed on the state and the extended shape of the unoccluded area, including:

[0032] The intersection of the occluded area boundary line and the target center line is calculated by using the radar position, the obstacle position, and one-step prediction of the target state and the extended shape;

[0033] The ratio of the unoccluded area to the overall target area size is calculated as the unoccluded proportion factor according to the intersection of the occluded area boundary line and the target center line and the target state and the extended shape;

[0034] The one-step prediction representation of the state and the extended shape of the unoccluded area is calculated by using the updated unoccluded proportion factor.

[0035] In the embodiment of the present application, the state and the extended shape of the unoccluded area are updated by using the random matrix method, and the unoccluded proportion factor is updated, including:

[0036] After receiving a new frame of measurements, the state of the unoccluded region of the target is updated based on the random matrix method using the measurements;

[0037] The extended shape of the unoccluded region of the target is updated based on the random matrix method using the measurements;

[0038] The updated unoccluded proportion factor is calculated using the updated unoccluded region extended shape and the overall target extended shape obtained by one-step prediction.

[0039] In the embodiments of the present application, the overall target is updated using the updated unoccluded proportion factor and the state and extended shape of the unoccluded region, including:

[0040] The overall target extended shape is updated using the updated unoccluded proportion factor and the updated unoccluded region extended shape;

[0041] The overall target state is updated using the updated unoccluded proportion factor and the updated unoccluded region state and extended shape.

[0042] In the embodiments of the present application, the overall target state is updated using the updated unoccluded proportion factor and the updated unoccluded region state and extended shape, including:

[0043] The elliptical shape of the overall target is calculated according to the updated overall target extended shape;

[0044] Two vertices of the unoccluded region in the elliptical shape representation are obtained from the state and extended shape of the unoccluded region;

[0045] The overall target state is obtained based on the rotation matrix, the major axis of the ellipse, and the vertices of the elliptical shape of the unoccluded region.

[0046] In another aspect of the present application, a target tracking device is provided, including:

[0047] The acquisition module is configured to acquire measurement data of the target and perform initialization;

[0048] The tracking module is configured to perform one-step prediction of the state and extended shape of the target track at the previous time, determine whether there is occlusion, calculate the one-step prediction representation of the state and extended shape of the unoccluded region using the unoccluded proportion factor, update the state and extended shape of the unoccluded region, and update the overall target using the updated unoccluded proportion factor and the state and extended shape of the unoccluded region.

[0049] The calculation module is configured to calculate and update the unoccluded proportion factor.

[0050] The present application has the following beneficial effects due to the above technical solutions:

[0051] The present application proposes a partial occlusion extended target tracking algorithm based on a random matrix method, which improves the accuracy of state and extended shape estimation under the condition that the target is occluded, and reduces the estimation error of the target state; the present application creatively proposes a state and extended shape estimation method suitable for the condition that the target is occluded based on the random matrix method, realizes extended target tracking under the condition of occlusion, and can still accurately estimate the motion state (position, speed, acceleration) and extended shape (size, orientation) of the target under the condition of occlusion.

[0052] The present application can be applied to the fields of intelligent transportation and automatic driving, and can accurately track the occluded target under the conditions that the target is occluded by an obstacle, the target is occluded by another target under the condition that multiple targets are tracked, and the target is tracked under the condition of crowded roads, and solves the problems of state and extended shape estimation errors under the condition that the target is occluded. BRIEF DESCRIPTION OF DRAWINGS

[0053] The accompanying drawings, which are included to provide a further understanding of the present application and constitute a part of this application, illustrate certain non-limiting embodiments of the present application and do not limit the present application, and in the drawings:

[0054] Figure 1 It is a flowchart of the partial occlusion extended target tracking algorithm based on the random matrix method.

[0055] Figure 2 It is a schematic diagram of the vertex under the elliptical representation of the target and the intersection point of the target central axis and the occlusion boundary.

[0056] Figure 3 It is a schematic diagram of the vertex under the elliptical representation of the unoccluded area and the overall target center point.

[0057] Figure 4 It is a simulation scene schematic diagram.

[0058] Figures 5(a)-5(e) It is a schematic diagram of the original algorithm result under the simulation scene.

[0059] Figures 6(a)-6(e) It is a schematic diagram of the improved algorithm result under the simulation scene.

[0060] Figure 7 It is a structural block diagram of the target tracking device. DETAILED DESCRIPTION

[0061] The present application will be described in detail below in combination with the drawings and specific embodiments, and the schematic embodiments and descriptions of the present application are used to explain the present application, but do not limit the present application.

[0062] As shown in the drawings, Figure 1 The present application provides a partial occlusion extended target tracking method based on a random matrix method, which includes the following steps:

[0063] S101: receive the measurement data at the initial time, initialize the target state and the extended form.

[0064] Specifically, the following steps are included:

[0065] 11) convert the sensor angle and distance measurement at the initial time in the polar coordinate system to the rectangular coordinate system;

[0066] 12) calculate the measurement point mean and dispersion matrix;

[0067] For n k measurement data at time k The measurement mean is The dispersion matrix is The calculation formula is:

[0068]

[0069]

[0070] The dispersion matrix is n k -1 times the covariance matrix;

[0071] 13) initialize the target state and the extended form using the measurement point mean and the dispersion matrix.

[0072] The state of the target at time k is x k = [x y v x v y ] T , x and y represent the x and y coordinates of the target center point in the Cartesian coordinate system, the covariance matrix of the state of the target at time k is P k , v x and v y represent the speed of the target center point in the x and y directions respectively; X k represents the extended form of the target at time k, which is a 2x2 matrix, v k represents the degree of freedom of the extended form at time k, the initial target state x0 and the extended form X0 are initialized as follows:

[0073]

[0074]

[0075] In addition, the initial state covariance matrix P0 and the degree of freedom v0 of the extended form are adjustable parameters, which need to be adjusted according to the actual situation of the radar. Preferably, a recommended amount is given, which can be adjusted on this basis:

[0076]

[0077] S102: One-step prediction of state and extended form of target track at last time is performed by using random matrix method.

[0078] Specifically, the method comprises the following steps:

[0079] 21) Obtain the motion model of the target, and perform one-step prediction of state and extended form of target track at last time by using random matrix method;

[0080] The motion model of the motion state of the target is as follows:

[0081]

[0082] In the formula, x k is the motion state at time k, Φ k is a state transition matrix, x k-1 is the motion state at time k-1, w k is process noise, is a Gaussian distribution with mean 0 and variance , D k is a covariance matrix of process noise in a one-dimensional physical space, is a Kronecker product, and X k is an extended form;

[0083] wherein, F k is a state transition matrix in a one-dimensional physical space, I d is a unit matrix, and d is the dimension of the physical space; is the variance of acceleration in a single-dimensional direction, is a parameter matrix;

[0084] Let denote a set composed of n k measurement data at time k, and let denote measurement information at time k-1 and before;

[0085] One-step prediction of the motion state is as follows:

[0086]

[0087]

[0088] In the formula, is a one-step prediction value of the state of the target at time k, is an estimated value of the motion state at time k-1, and P k|k-1F is the one-step prediction of the covariance of the target state at time k k is the state transition matrix in one-dimensional physical space

[0089] 22) The last time's track hypothesis is extended by one step using the random matrix method; the extended shape evolution model describes and characterizes the change of the target's extended shape over time

[0090]

[0091] wherein X k is the extended shape at time k, X k-1 is the extended shape at time k-1, is a Wishart distribution, δ k is the degree of freedom of the evolution distribution, A k is the extended shape evolution matrix

[0092] The one-step prediction of the extended shape is

[0093]

[0094]

[0095]

[0096] wherein: is the one-step prediction of the target's extended shape at time k, is the estimate of the target's extended shape at time k-1, is the one-step prediction of the target's extended shape degree of freedom at time k, is the estimate of the target's extended shape degree of freedom at time k-1, d is the dimension of the physical space, λ k-1 is an intermediate variable

[0097] S103: Determine whether there is an occlusion using the radar position and the obstacle position and shape.

[0098] Specifically, the following steps are included:

[0099] 31) Calculate the occluded area range using the radar position and the obstacle position;

[0100] The radar position is defined as Pos radar = [x r y r ] T , the radar pointing angle angle pointer is defined as the middle angle of the radar detection range angle, and the vertex set of the obstacle is wherein n bThe total number of obstacle vertexes, the maximum anticlockwise angle φ relative to the radar pointing angle can be calculated by calculating the difference between the angle of the line connecting each vertex and the radar position and the radar pointing angle b and the maximum clockwise angle θ b , through the pointing angle angle pointer , the maximum anticlockwise angle φ b and the maximum clockwise angle θ b can represent the range of the occlusion area, which is used for subsequent occlusion judgment.

[0101] 32) Determine whether there is an occlusion according to the occlusion area range and the one-step prediction value of the target state and the extended form.

[0102] The specific steps are as follows:

[0103] 321) Calculate the four vertices of the target according to the target state and the one-step prediction value of the extended form and , wherein the method for obtaining the target elliptical shape matrix representation based on is as follows:

[0104]

[0105]

[0106] wherein is obtained from the inverse Wishart distribution property, representing the elliptical shape of the target, Z k-1 is the measurement set at k-1 moment, is the one-step prediction value of the degree of freedom, d represents the dimension of the physical space, R k|k-1 is a rotation matrix, D k|k-1 is a diagonal matrix, the upper left element is the square of the semi-major axis a k|k-1 of the ellipse, and the lower right element is the square of the semi-minor axis b k|k-1 of the ellipse, and the semi-major axis a k|k-1 of the ellipse, the semi-minor axis b k|k-1 , the rotation matrix R k|k-1 , and the target center point can be directly obtained

[0107] 322) According to the vertex set of the target wherein n t =4, i.e. there are four vertices, and the radar position and the radar pointing angle, the maximum anticlockwise angle φ relative to the radar pointing angle t and the maximum clockwise angle θ t can be obtained;

[0108] 323) the maximum angle φ counter-clockwise b and the maximum angle θ clockwise b , the maximum angle φ counter-clockwise t and the maximum angle θ clockwise t It can be determined whether there is an occlusion:

[0109] (1) (φ t < φ b and φ t > θ b ) or (θ t < φ b and θ t > θ b ): there is an occlusion

[0110] (2) (φ t < φ b and φ t > θ b ) and (θ t < φ b and θ t > θ b ): full occlusion

[0111] According to the above determination condition, it can be determined whether the target is occluded and whether it is fully occluded or partially occluded. If condition (1) is not met, the target is not occluded, and the original random matrix algorithm is used for estimation. If condition (2) is met, the target is fully occluded and no operation is performed. If condition (1) is met but condition (2) is not met, the target is partially occluded and subsequent methods can be used for estimation.

[0112] S104: If there is an occlusion, calculate the unoccluded proportion factor, and make one-step prediction of the state and extended form of the unoccluded area.

[0113] Specifically, the following steps are included:

[0114] 41) Calculate the intersection of the occlusion area boundary line and the target center line using the radar position, obstacle position, and one-step prediction of the state and extended form of the target;

[0115] 42) Calculate the ratio of the unoccluded area to the overall target area size as the unoccluded proportion factor based on the intersection of the occlusion area boundary line and the target center line and the state and extended form of the target.

[0116] As shown in FIG. 4, the intersection of the occlusion area boundary line and the target center line is Figure 2 of the two vertices of the target center line, the one inside the occlusion area is defined as and the one outside the occlusion area is defined as ​The unoccluded scale factor is then calculated as follows:

[0117]

[0118] where the function dis(p1, p2) denotes the distance between two points, a p,k|k-1 denotes the long semi-axis of the ellipse calculated from the random matrix representation of the unoccluded region's extended shape, a k|k-1 denotes the long semi-axis of the ellipse calculated from the random matrix representation of the overall target's extended shape.

[0119] 43) Calculate the one-step prediction representation of the unoccluded region's state and extended shape using the unoccluded scale factor.

[0120] Specifically includes the following steps:

[0121] 431) The state of the unoccluded region is defined as The velocity component of this vector, i.e. the last two terms, is still the same as The first two terms of this vector, i.e. the position components, are equal to The midpoint of the two points;

[0122] 432) The extended shape of the unoccluded region is defined as The calculation of this extended shape matrix needs to be based on The elliptical extended shape representation of the target is obtained, which has been described in step 321), and the formula is given again here:

[0123]

[0124]

[0125] Can be calculated as follows:

[0126]

[0127]

[0128] where R p,k|k-1 is the rotation matrix corresponding to the elliptical extended shape of the unoccluded region, and R p,k|k-1 = R k|k-1 .

[0129] S105: Receive the measurement data at the current time, update the state and extended shape of the unoccluded region using the random matrix method, and update the unoccluded scale factor.

[0130] Specifically includes the following steps:

[0131] 51) After receiving a new frame of measurements, the state of the unoccluded region of the target is updated based on the random matrix method using the measurement points;

[0132] Let the measurement set at time k be n k The number of measurements at time k, under the framework of random matrix, the following measurement model structure is given:

[0133]

[0134] In the formula, is the measurement, is the measurement matrix of the system, x p,k is the unoccluded region motion state at time k, is an independent Gaussian white noise;

[0135] The update of the unoccluded region motion state is:

[0136]

[0137] In the formula, is the estimated value of the unoccluded region motion state at time k, is a one-step prediction value of the unoccluded region target state at time k, K k is the filter gain, I d is the unit matrix, is the Kronecker product, G k is the estimation error, P k is the estimated value of the covariance of the unoccluded region motion state at time k, P k|k-1 is a one-step prediction value of the covariance of the unoccluded region target state at time k, S k is the innovation covariance matrix;

[0138] Wherein, the innovation covariance matrix S k is calculated as follows:

[0139]

[0140] The filter gain K k is calculated as follows:

[0141]

[0142] The estimation error G k is calculated as follows:

[0143]

[0144] In the formula, is the mean of the measurement data at time k;

[0145] 52) update the extended shape of the unoccluded region of the target based on the random matrix method using the measurement points;

[0146] Based on the measurement model structure, the update of the extended shape can be obtained as:

[0147]

[0148] where, is the estimate of the extended shape of the target at time k, is the one-step prediction of the extended shape of the target at time k, B k is the extended shape observation matrix, is the scatter matrix of the measurement data at time k, n k is the number of measurements at time k, is the estimate of the degrees of freedom of the extended shape of the target at time k, is the one-step prediction of the degrees of freedom of the extended shape of the target at time k;

[0149] 53) calculate the updated unoccluded fraction using the updated unoccluded region extended shape and the overall target extended shape obtained from the one-step prediction.

[0150] The updated unoccluded region extended shape is represented as Based on the elliptical shape representation of the unoccluded region can be obtained, similar to step 321):

[0151]

[0152]

[0153] where, is the updated degrees of freedom, R p,k is the rotation matrix, D p,k is the diagonal matrix, a p,k ,b p,k are the long and short semi-axes of the ellipse, respectively, then the updated unoccluded fraction τ k is calculated as follows:

[0154]

[0155] S106: update the overall target using the updated unoccluded fraction and the state and extended shape of the unoccluded region.

[0156] Specifically, the following steps are included:

[0157] 61) update the overall target extended shape using the updated unoccluded fraction and the updated unoccluded region extended shape;

[0158] Overall target extended shape X k The evolution model is represented as follows:

[0159]

[0160] A' k = A k R p,k T k R p,k T

[0161]

[0162] where, is the extended shape degree of freedom updated in step 52). Since the overall target extended shape X k is still consistent with the Wishart distribution under the condition that the partial extended shape X p,k is known, the random matrix method can still be used to estimate the overall target extended shape.

[0163] The estimation of the extended shape is:

[0164]

[0165]

[0166]

[0167] where, is the overall target extended shape at time k, is the target extended shape of the unoccluded area, is the degree of freedom of the overall target extended shape, is the degree of freedom of the overall target extended shape, d is the dimension of the physical space, λ p,k is an intermediate variable.

[0168] 62) Update the overall target state using the updated unoccluded proportion factor and the updated unoccluded area state and extended shape.

[0169] 621) Calculate the elliptical shape representation of the overall target according to the updated overall target extended shape

[0170]

[0171]

[0172] where, is the degree of freedom, R​k is a rotation matrix, D k is a diagonal matrix, a k ,b k are the long and short semi-axes of the ellipse respectively;

[0173] 622) As shown in the following equation, the state Figure 3 and the spread form of the unoccluded region can be obtained. Two vertices of the unoccluded region in the elliptical shape representation can be obtained, where is close to the occluded region, and is far away from the occluded region.

[0174] 623) Based on the rotation matrix R k , the long semi-axis a k of the ellipse, and the vertices of the elliptical shape of the unoccluded region , the state of the overall target can be obtained and represented as:

[0175]

[0176] S107: Output the target track.

[0177] Specifically, it includes:

[0178] The tracking result of the target motion state at time k is

[0179] According to the properties of the inverse Wishart distribution, we can obtain which is the elliptical shape representation of the target spread form at time k.

[0180] The present application will be further described below through specific embodiments.

[0181] S101: Receive the millimeter wave radar measurement at the starting time, and convert it to the sensor rectangular coordinate system with the millimeter wave radar as the origin, and then convert it to the absolute rectangular coordinate system based on the position and pose of the sensor as appropriate; calculate the mean and dispersion matrix of the measurement points, and initialize the target state and spread form using the mean and dispersion matrix;

[0182] S102: After initialization, according to the motion model and the spread form evolution model, the state and spread form of the target at the last time are predicted by one step using the random matrix method, and the state prediction parameters P k|k-1 and the spread form prediction parameters

[0183] S103: Calculate the occluded region range using the radar position and the obstacle position, and the radar pointing angle anglepointer the maximum angle φ clockwise b and the maximum angle θ counterclockwise b The range of the occluded area can be represented, and then it is determined whether the target is occluded and whether it is partially occluded;

[0184] S104: If there is partial occlusion, the unoccluded ratio factor τ is calculated by calculating the intersection of the occluded area boundary line and the target center line k|k-1 , and the state and extension form of the unoccluded area are predicted one step to obtain the one-step prediction of the state and the extension form of the unoccluded area

[0185] S105: The state and extension form of the unoccluded area are updated using the random matrix method to obtain the updated state and the extension form of the unoccluded area Then, the unoccluded ratio factor is updated using the updated state and extension form of the unoccluded area to obtain the updated unoccluded ratio factor τ k ;

[0186] S106: The updated unoccluded ratio factor and the state and extension form of the unoccluded area are used to construct the evolution model of the overall target state under the condition that the state of the unoccluded area is known S106: The updated unoccluded ratio factor and the state and extension form of the unoccluded area are used to construct the evolution model of the overall target state under the condition that the state of the unoccluded area is known and the extension form of the overall target are updated using the random matrix method to obtain the state and the extension form of the overall target

[0187] S107: The updated target state, extension form, and elliptical shape representation of the extension form are output.

[0188] Figure 4 In the simulation scenario shown, the millimeter wave radar is stationary, located at x=12m, y=0.85m, the obstacle is a rectangle with four vertices (12.9, 3.2), (11.1, 3.2), (11.1, -1.5), and (12.9, -1.5), the extended target is 6m long and 1.5m wide, the target moves in the radar field of view, slowly approaches the occluded area from a distance away from the occluded area, and slowly enters the occluded area, the target movement speed is about 1m / s, the radar measurement noise is subject to a Gaussian distribution with a mean of 0 and a covariance matrix R k = diag([0.5 2 , 0.5 2 ])m 2 , and the sampling interval T=0.1s.

[0189] Figures 5(a)-5(e)The result of using the original random matrix based extended target tracking method in the simulation scene is shown. When the target just enters the occluded area at step 110, the method has little deviation in the estimation of the extended shape and state of the target, and when the method estimates the extended shape and state of the target at step 130, a great deviation has occurred, the estimated target position is inaccurate, and the extended shape is also greatly reduced compared with the true value, and when the method estimates the extended shape at step 140, the estimated extended shape is close to half of the true shape, and a great deviation has occurred.

[0190] Figures 6(a)-6(e) The result of using the random matrix based partial occlusion extended target tracking algorithm in the simulation scene is shown. The algorithm can always guarantee good estimation effect of the state and extended shape of the target when the target enters the occluded area, and the estimated target almost coincides with the true target at step 135, and the estimation accuracy is high, and the algorithm still has good estimation effect of the state and extended shape of the target at step 140 without the radar measurement point of the target.

[0191] Please refer to Figure 7 The target tracking device 100 provided by the embodiment of the application is shown, and the target tracking device 100 comprises:

[0192] The acquisition module 110 is configured to acquire measurement data of a target and perform initialization.

[0193] The tracking module 120 is configured to perform one-step prediction of the state and extended shape of the target track at the last moment, judge whether there is occlusion, calculate one-step prediction expression of the state and extended shape of the un-occluded area by using an un-occluded proportion factor, update the state and extended shape of the un-occluded area, and update the whole target by using the updated un-occluded proportion factor and the state and extended shape of the un-occluded area.

[0194] The calculation module 130 is configured to calculate and update the un-occluded proportion factor.

[0195] Optionally, the acquisition module 110 acquires measurement data at each moment and performs coordinate conversion to obtain n k measurement data at the kth moment and initializes the target state x0 and the extended shape X0 by using the measurement data at the first moment.

[0196] Optionally, the tracking module 120 uses the random matrix method to perform one-step prediction of the state and the extended shape of the target track at the last time, uses the radar position and the obstacle position and shape to determine whether there is an occlusion, uses the un-occluded proportion factor to calculate the one-step prediction representation of the state and the extended shape of the un-occluded area, receives the measurement data at the current time, uses the random matrix method to update the state and the extended shape of the un-occluded area, and uses the updated un-occluded proportion factor and the state and the extended shape of the un-occluded area to update the overall target.

[0197] Optionally, the calculation module 130 calculates and updates the un-occluded proportion factor, calculates the ratio of the size of the un-occluded area to the overall target area according to the intersection of the occluded area boundary line and the target center line and the state and the extended shape of the target, and obtains the un-occluded proportion factor, and uses the updated un-occluded area extended shape and the overall target extended shape obtained by one-step prediction to calculate the updated un-occluded proportion factor.

[0198] The above experimental results verify that the application improves the state and the extended shape estimation effect of the partially occluded target, realizes accurate tracking of the target when the target slowly enters the occluded area, and has great theoretical and practical value.

[0199] The application is not limited to the above embodiments, and based on the technical solutions disclosed in the application, those skilled in the art can make some substitutions and deformations to some technical features without creative labor according to the disclosed technical content, and these substitutions and deformations are all within the protection scope of the application.

Claims

1. A method for tracking partially occluded extended targets based on the random matrix method, characterized in that, include: Receive the initial measurement data and initialize the target state and extended form; The state and extended shape of the target trajectory at the previous time step are predicted in one step using the random matrix method. Use radar position and obstacle position to determine if there is any obstruction; If occlusion exists, calculate the unoccluded ratio factor and make a one-step prediction of the state and expansion pattern of the unoccluded area; Receive the measurement data at the current moment, update the state and expansion pattern of the unobstructed area using the random matrix method, and update the unobstructed scale factor; The overall target is updated using the updated unoccluded scale factor and the state and expansion shape of the unoccluded area; Output the target trajectory; Calculate the unobstructed scale factor to predict the state and expansion pattern of the unobstructed area in one step, including: The intersection of the boundary line of the obstruction area and the center line of the target is calculated in one step using radar position, obstacle position, and the state and expansion pattern of the target; The ratio of the size of the unobstructed area to the size of the overall target area is calculated based on the intersection of the boundary line of the obstructed area and the center line of the target, as well as the state and expansion shape of the target. This ratio is used as the unobstructed scale factor. The state and expansion morphology of the unoccluded area are predicted and represented in one step by using the unoccluded scale factor; The overall target is updated using the updated unoccluded scale factor and the state and expansion shape of the unoccluded area, including: The overall target expansion shape is updated using the updated unoccluded scale factor and the updated unoccluded area expansion shape; The overall target state is updated using the updated unoccluded scale factor, the updated unoccluded area state, and the expanded shape. The overall target state is updated using the updated unoccluded scale factor, updated unoccluded region state, and expanded shape, including: Calculate the elliptical shape of the overall target based on the updated overall target expansion shape; The two vertices of the elliptical shape representation of the unoccluded region are derived from the state and expanded shape of the unoccluded region; the vertices closer to the occluded region are... The area far from the obstruction is ; According to the rotation matrix semi-major axis of the ellipse The elliptical shape of the unobstructed area's vertices To obtain the overall goal status: 。 2. The method according to claim 1, characterized in that, Receive initial measurement data, initialize the target state and extended form, including: Transform the sensor measurements from the polar coordinate system at the initial moment to the rectangular coordinate system; The target state and extended form are initialized using the mean and divergence matrix of the measurements.

3. The method according to claim 1, characterized in that, One-step prediction of the state and extended morphology of the target trajectory at the previous time step using the stochastic matrix method includes: Obtain the target's motion model, and use the random matrix method and the motion model to make a one-step state prediction of the target's trajectory at the previous time step; The extended shape of the target trajectory at the previous time step is predicted in one step using the stochastic matrix method. The extended shape evolution model describes and characterizes the extended shape that changes over time, and the extended shape is predicted in one step.

4. The method according to claim 1, characterized in that, Determining whether there is obstruction using radar position and obstacle position includes: The extent of the obstruction area is calculated using the radar position and the obstacle position. By calculating the difference between the angle of the line connecting each obstacle vertex to the radar position and the radar pointing angle, the maximum counterclockwise angle and the maximum clockwise angle relative to the radar pointing angle are calculated. The range of the obstructed area is represented by the pointing angle, the maximum counterclockwise angle, and the maximum clockwise angle. The presence of occlusion is determined by a one-step prediction value based on the occlusion area, target status, and expansion pattern.

5. The method according to claim 4, characterized in that, Determining whether occlusion exists based on the one-step prediction value of the occlusion area range, target state, and expansion pattern, including: Calculate the four vertices of the target based on the one-step prediction value of the target state and extended form; Based on the target's vertex set, radar position, and radar pointing angle, the maximum counterclockwise and maximum clockwise angles of the target relative to the radar pointing angle are obtained. Based on the maximum counterclockwise and maximum clockwise angles of the obstacle, and the maximum counterclockwise and maximum clockwise angles of the target relative to the radar pointing angle, determine whether there is any obstruction.

6. The method according to claim 1, characterized in that, The state and expansion morphology of the unoccluded region are updated using the random matrix method, and the unoccluded scale factor is updated, including: After receiving a new frame of measurement, the state of the unoccluded area of ​​the target is updated using the measurement points based on the random matrix method; The extended shape of the unobstructed area of ​​the target is updated using measurement points based on the random matrix method; The updated unoccluded area expansion morphology and the overall target expansion morphology obtained from one-step prediction are used to calculate the updated unoccluded scale factor.

7. A target tracking apparatus for implementing the method as described in any one of claims 1-6, characterized in that, include: The acquisition module is used to acquire and initialize the target's measurement data; The tracking module is used to predict the state and extended shape of the target track in one step from the previous moment and to determine whether there is occlusion. The state and expansion morphology of the unoccluded area are predicted in one step using the unoccluded scale factor; the state and expansion morphology of the unoccluded area are updated; and the overall target is updated using the updated unoccluded scale factor and the state and expansion morphology of the unoccluded area. The calculation module is used to calculate and update the unshaded scale factor.

Images