Target tracking method based on Gaussian process and joint probability data association

Abstract

The invention belongs to the technical field of multi-target tracking, and particularly relates to a target tracking method based on Gaussian process and joint probability data association, which comprises the following steps: S1, obtaining state estimation of a current frame, and representing the current frame by k; s2, predicting state estimation of a next frame through the state transition matrix, wherein the next frame is represented by k + 1; s3, constructing a joint tracking gate based on a Gaussian process to select a next frame of effective measurement for updating a target state; s4, summarizing the source condition of each effective measurement of the next frame, and obtaining related events about measurement sources; and S5, taking approximate statistics of effective measurement of related events and all frames as conditions. According to the method, the Gaussian process and the probability data association algorithm are combined, so that the precision of target shape estimation in the clutter environment is greatly improved, more accurate and abundant target information can be provided, detection and recognition of the target are facilitated, and meanwhile, the actual position of the target can be determined more accurately.

Description

Technical Field

[0001] The invention belongs to the technical field of multi-target tracking, and in particular relates to a target tracking method based on Gaussian process and joint probability data association. Background Art

[0002] Target tracking is an important research direction in the field of computer vision. It mainly involves feature extraction and modeling of specific targets in image sequences, and using these models to obtain or mark the position information of the target in the dynamic image sequence to achieve tracking.

[0003] The basic purpose of target tracking is to estimate the trajectory of an object in the image plane, that is, to assign a consistent label to the same target in different frames of a video. Specifically, the target tracking system identifies and tracks the position, shape, motion and other information of a specific target in the scene by analyzing a continuous sequence of images. The basic framework of online target tracking includes feature extraction, motion model, appearance model and online update mechanism. Feature extraction involves the image features of the target, such as grayscale, color, texture, Haar rectangle features, directional gradient histogram and depth features. The motion model is used to generate candidate samples, describe the relationship between the target motion states, and predict the target image area. The appearance model is used to confirm whether the candidate image area is the target area, and locate the spatial position of the target by extracting the image features of the area and screening or matching.

[0004] The application of target tracking is very extensive, including but not limited to security fields (such as vehicle tracking and license plate recognition), monitoring fields (such as face recognition and gait recognition), inspection fields (such as drone tracking and robot navigation), as well as smart life (such as human-computer interaction and VR / AR), smart cities (such as traffic monitoring) and smart industries (such as telemedicine). These applications mainly involve the acquisition of the spatial position, shape and size of any target in a video or continuous semantically related images. As a supplement to the detection algorithm, target tracking can reduce the complexity of the entire system. However, the current tracking algorithms have high computational complexity and long time consumption in complex scenarios (such as other noise and clutter), making them difficult to apply in real-time systems.

[0005] Therefore, in response to the above technical problems, it is necessary to provide a target tracking method based on Gaussian process and joint probability data association.

[0006] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0007] The object of the present invention is to provide a target tracking method based on Gaussian process and joint probability data association, which can solve the problem of multi-target tracking in a clutter environment.

[0008] In order to achieve the above object, a technical solution provided by a specific embodiment of the present invention is as follows:

[0009] A target tracking method based on Gaussian process and joint probability data association includes the following steps:

[0010] S1, obtain the current frame state estimation, the current frame is represented by k;

[0011] S2, predict the next frame state estimate through the state transfer matrix, and the next frame is represented by k+1;

[0012] S3, construct a joint tracking gate based on Gaussian process to select the next frame of valid measurement for updating the target state;

[0013] S4, summarizing the sources of each valid measurement in the next frame, and obtaining relevant events about the measurement sources;

[0014] S5, taking the relevant events and the approximate statistics of the valid measurements of all frames as conditions, obtaining the state estimation corresponding to the relevant events based on the Kalman filtering method;

[0015] S6. Based on the Bayesian probability formula, the weight of the relevant events is obtained with the valid measurements of all frames as the condition;

[0016] S7. Based on the overall probability formula, the state estimates of all relevant events and the corresponding event weights are summarized to obtain the fused next frame state estimate.

[0017] In one or more embodiments of the present invention, in S1, the current frame state estimation includes a state estimation and covariance P k , state estimation The calculation formula is as follows:

[0018]

[0019] in, represents the estimated motion state of the target center point, Represents the target contour state estimate.

[0020] In one or more embodiments of the present invention, the The calculation formula is:

[0021]

[0022] Among them, [x k ,yk ] is the target position vector, is the target velocity vector, represents the target heading angle, Indicates the target heading angular rate.

[0023] In one or more embodiments of the present invention, in S2, the state transfer matrix F k The formula is:

[0024]

[0025] Among them, ℉ k represents the target motion state transfer matrix, Represents the target contour state transition matrix.

[0026] In one or more embodiments of the present invention, the next frame state estimation includes a predicted state, a predicted covariance, and a predicted measurement, wherein the calculation formulas of the predicted state and the predicted covariance are respectively as follows:

[0027]

[0028] in, represents the predicted state of the target in the next frame, P k+1|k represents the prediction covariance, θ k Represents the process noise covariance of the previous frame.

[0029] In one or more embodiments of the present invention, the prediction measurement The calculation formula is:

[0030]

[0031] in, represents the target center position predicted for the next frame, μ s is the expected scale factor, To predict the rotation factors, is the predicted Gaussian process coefficient for the next frame, The object contour state predicted for the next frame; and Respectively represent the angle of the jth current measurement of the next frame relative to the predicted target center position in the local coordinate system and the global coordinate system, Represents the predicted target heading angle.

[0032] In one or more embodiments of the present invention, S3 includes: the j-th predicted measurement noise covariance and innovation covariance of the next frame The solution steps are:

[0033]

[0034] in, represents the scale factor covariance, R is the measurement noise covariance, and denote the j-th predicted measurement noise covariance and the predicted profile measurement noise covariance of the next frame, σ(·,·) denotes the square exponential covariance function (SE), ∑(·,·) denotes the square exponential covariance function matrix, Represents the j-th Jacobian matrix of the next frame.

[0035] In one or more embodiments of the present invention, in S4, the related events are used indicates that, and The solution formula is:

[0036]

[0037] Among them, m k+1 Indicates the number of valid measurements obtained in the next frame, Indicates the number of measurements of the target, n φ Indicates that when the number of measurements from the target is Time related events The number of Of the valid measurements, one comes from the target All valid measurements are clutter n φ =1.

[0038] In one or more embodiments of the present invention, in S6, the solution formula for the relevant event weight is:

[0039]

[0040] In one or more embodiments of the present invention, in S7, the next frame state estimation includes state estimation and covariance estimation, and the calculation formula is as follows:

[0041]

[0042] in, Based on related events The state estimate of is the corresponding Kalman threshold, Indicates related events The joint information in Indicates related events The joint Jacobian matrix in Related events The joint measurement error in when hour,

[0043] Compared with the prior art, the target tracking method based on Gaussian process and joint probabilistic data association of the present invention greatly improves the accuracy of target shape estimation in a clutter environment by combining Gaussian process with probabilistic data association algorithm, thereby providing more accurate and rich target information, which is beneficial to target detection and identification. At the same time, it can also more accurately determine the actual position of the target. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 The present invention is a flowchart of a target tracking method based on Gaussian process and joint probability data association in one embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0047] like Figure 1 As shown, a target tracking method based on Gaussian process and joint probability data association in one embodiment of the present invention includes the following steps:

[0048] S1, obtain the current frame state estimation, the current frame is represented by k;

[0049] S2, predict the next frame state estimate through the state transfer matrix, and the next frame is represented by k+1;

[0050] S3, construct a joint tracking gate based on Gaussian process to select the next frame of valid measurement for updating the target state;

[0051] S4, summarizing the sources of each valid measurement in the next frame, and obtaining relevant events about the measurement sources;

[0052] S5, taking the relevant events and the approximate statistics of the valid measurements of all frames as conditions, obtaining the state estimation corresponding to the relevant events based on the Kalman filtering method;

[0053] S6. Based on the Bayesian probability formula, the weight of the relevant events is obtained with the valid measurements of all frames as the condition;

[0054] S7. Based on the overall probability formula, the state estimates of all relevant events and the corresponding event weights are summarized to obtain the fused next frame state estimate.

[0055] In S1, the current frame state estimation includes the state estimation and covariance P k , state estimation The calculation formula is as follows:

[0056]

[0057] in, represents the estimated motion state of the target center point, Represents the target contour state estimate.

[0058] in addition, The calculation formula is:

[0059]

[0060] Among them, [x k ,y k ] is the target position vector, is the target velocity vector, represents the target heading angle, Indicates the target heading angular rate.

[0061] In S2, the state transfer matrix F k The formula is:

[0062]

[0063] Among them, ℉ k represents the target motion state transfer matrix, Represents the target contour state transition matrix.

[0064] In addition, the next frame state estimation includes predicted state, predicted covariance and predicted measurement, where the calculation formulas of predicted state and predicted covariance are as follows:

[0065]

[0066] in, represents the predicted state of the next frame, P k+1|k represents the prediction covariance of the next frame, θk Represents the process noise covariance of the previous frame.

[0067] Specifically, the prediction measurement The calculation formula is:

[0068]

[0069] in, represents the target center position predicted for the next frame, μ s is the expected scale factor, To predict the rotation factors, is the predicted Gaussian process coefficient for the next frame, The object contour state predicted for the next frame; and Respectively represent the angle of the jth current measurement of the next frame relative to the predicted target center position in the local coordinate system and the global coordinate system, Represents the predicted target heading angle.

[0070] In S3, it specifically includes the j-th predicted measurement noise covariance and the innovation covariance of the next frame The solution steps are:

[0071]

[0072] in, represents the scale factor covariance, R is the measurement noise covariance, and denote the j-th predicted measurement noise covariance and the predicted profile measurement noise covariance of the next frame, σ(·,·) denotes the square exponential covariance function (SE), ∑(·,·) denotes the square exponential covariance function matrix, Represents the j-th Jacobian matrix of the next frame.

[0073] In S4, the relevant events are represented by Represents, and the solution formula for related events is:

[0074]

[0075] Among them, m k+1 Indicates the number of valid measurements obtained in the next frame, represents the number of measurements of the target, Indicates that when the number of measurements from the target is Time related events The number of Of the valid measurements, one is from the target All valid measurements are clutter

[0076] In S5, the solution formula for the relevant event state estimation is as follows:

[0077]

[0078] in, Based on related events The state estimate of is the corresponding Kalman threshold, Indicates related events The joint information in Indicates related events The joint Jacobian matrix in Related events The joint measurement error in

[0079] In S6, the number of clutter follows Poisson distribution and is evenly distributed within the detection range. The detection probability of the target is P D , based on the effective measurement at the current moment and the approximate statistics of all measurements at the past moments, the weights of related events are obtained, namely:

[0080]

[0081] In S7, the overall probability formula is used to calculate the conditional estimates of all relevant events. With the corresponding event weight To summarize, we can get the next frame state estimation and covariance estimation, namely:

[0082]

[0083]

[0084] Among them, when hour,

[0085] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0086] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A target tracking method based on Gaussian process and joint probability data association, characterized in that: The following steps are involved: S1, obtain the current frame state estimation, the current frame is represented by k; S2, predict the next frame state estimate through the state transfer matrix, and the next frame is represented by k+1; S3, construct a joint tracking gate based on Gaussian process to select the next frame of valid measurement for updating the target state; S4, summarizing the sources of each valid measurement in the next frame, and obtaining relevant events about the measurement sources; S5, taking the relevant events and the approximate statistics of the valid measurements of all frames as conditions, obtaining the state estimation corresponding to the relevant events based on the Kalman filtering method; S6. Based on the Bayesian probability formula, the weight of the relevant events is obtained with the valid measurements of all frames as the condition; S7. Based on the overall probability formula, the state estimates of all relevant events and the corresponding event weights are summarized to obtain the fused next frame state estimate.

2. A target tracking method based on Gaussian process and joint probability data association according to claim 1, characterized in that: In S1, the current frame state estimation includes a state estimation and covariance P k , state estimation The calculation formula is as follows: in, represents the estimated motion state of the target center point, Represents the target contour state estimate.

3. The target tracking method based on Gaussian process and joint probability data association according to claim 2 is characterized in that: Said The calculation formula is: Among them, [x k ,y k ] is the target position vector, is the target velocity vector, represents the target heading angle, Indicates the target heading angular rate.

4. The target tracking method based on Gaussian process and joint probability data association according to claim 3 is characterized in that: In S2, the state transfer matrix F k The formula is: Among them, ℉ k represents the target motion state transfer matrix, Represents the target contour state transition matrix.

5. The target tracking method based on Gaussian process and joint probability data association according to claim 4, characterized in that: The next frame state estimation includes predicted state, predicted covariance and predicted measurement, wherein the calculation formulas of predicted state and predicted covariance are as follows: in, represents the predicted state of the target in the next frame, P k+1|k represents the prediction covariance of the next frame, θ k Represents the process noise covariance of the previous frame.

6. A target tracking method based on Gaussian process and joint probability data association according to claim 5, characterized in that: The predicted measurement The calculation formula is: in, represents the target center position predicted for the next frame, μ s is the expected scale factor, To predict the rotation factors, is the predicted Gaussian process coefficient for the next frame, The object contour state predicted for the next frame; and Respectively represent the angle of the jth current measurement of the next frame relative to the predicted target center position in the local coordinate system and the global coordinate system, Represents the predicted target heading angle.

7. The target tracking method based on Gaussian process and joint probability data association according to claim 6, characterized in that: S3 includes: the j-th predicted measurement noise covariance and the new information covariance of the next frame The solution steps are: in, represents the scale factor covariance, R is the measurement noise covariance, and denote the j-th predicted measurement noise covariance and the predicted profile measurement noise covariance of the next frame, σ(·,·) denotes the square exponential covariance function (SE), ∑(·,·) denotes the square exponential covariance function matrix, Represents the j-th Jacobian matrix of the next frame.

8. The target tracking method based on Gaussian process and joint probability data association according to claim 7, characterized in that: In S4, the relevant events are represented by indicates that, and The solution formula is: Among them, m k+1 Indicates the number of valid measurements obtained in the next frame, represents the number of measurements of the target, Indicates that when the number of measurements from the target is Time related events The number of Of the valid measurements, one comes from the target All valid measurements are clutter 9. The target tracking method based on Gaussian process and joint probability data association according to claim 8, characterized in that: In S6, the solution formula for the relevant event weight is:

10. The target tracking method based on Gaussian process and joint probability data association according to claim 9, characterized in that: In S7, the next frame state estimation includes state estimation and covariance estimation, and the calculation formula is as follows: in, Based on related events The state estimate of is the corresponding Kalman threshold, Indicates related events The joint information in Indicates related events The joint Jacobian matrix in Related events The joint measurement error in when hour,

Images