A radar target stable tracking method in a shielding environment

Abstract

The application discloses a radar target stable tracking method under a shielding environment, and comprises the following steps: a northeast two-dimensional coordinate system is established, and a target ship state is mapped into a state vector; a Kalman filter is used for prediction, a residual error and an innovation covariance matrix are calculated in combination with actual observation, and an elliptical threshold is constructed; if the observation falls into the threshold, the state is updated; otherwise, the observation is judged as a mismatched observation, and short-time missed detection prediction is performed; a sliding time window is set, when the number of mismatched observations in the window reaches a threshold, shielding judgment is performed in combination with the geometric relationship among the predicted positions of the ship and the target and the boundary of a static obstacle; if the condition is met, a shielding maintaining mode is entered, and the target track is continuously estimated by adjusting filter parameters or a predicted track; when a plurality of continuous observation frames fall into the threshold and the geometric shielding is removed, normal tracking is recovered through smooth transition. The application can effectively solve the problem of tracking interruption caused by the shielding of a radar target, and improve the continuity and stability of tracking under a shielding environment.

Description

Technical Field

[0001] This invention relates to the field of radar target tracking, and more particularly to a method for stable radar target tracking under obstructed conditions. Background Technology

[0002] Radar target tracking technology has wide applications in fields such as autonomous driving, ship navigation, and traffic monitoring. In real-world complex environments, there are often static or dynamic obstacles (such as buildings, bridges, large vehicles, mountains, etc.) between radar sensors and targets, which can cause radar beams to be blocked, resulting in intermittent target echoes and loss or mismatch of observations. Traditional radar target tracking methods usually assume that observation data is continuously available. When blockage occurs, target observations are missing for a long time, which can easily lead to filter divergence, trajectory interruption, or target loss.

[0003] Currently, common radar target tracking methods are mainly based on recursive Bayesian estimation frameworks such as Kalman filtering and particle filtering. For example, standard Kalman filtering achieves state estimation through two steps: prediction and update. When the residual between the observed and predicted values ​​exceeds a preset threshold, the nearest neighbor method or probabilistic data association is often used for observation matching. However, these methods have significant shortcomings in occlusion environments: First, the threshold region is fixed or based on simple statistical assumptions, making it difficult to adapt to sudden changes in observations caused by dynamic occlusion. Second, the handling of continuous mismatched observations is simple, usually directly discarding or maintaining the predicted values, lacking active discrimination and mode switching of occlusion states. Third, geometric relationships (such as the relative relationship between the ship's position, the predicted target position, and the obstacle boundary) are not used to help determine the timing of occlusion occurrence and resolution, resulting in large state jumps or slow convergence when resuming tracking after a brief occlusion. Summary of the Invention

[0004] This invention provides a stable radar target tracking method under obstructed environments to overcome the problems of observational abrupt changes caused by dynamic obstruction in the threshold region, the simplistic handling of continuous mismatched observations, and the large state jumps or slow convergence during tracking recovery.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: A method for stable radar target tracking in obstructed environments includes: S1. Establish a two-dimensional Cartesian coordinate system in the northeast direction, and map the motion state of the target ship observed by the radar to the two-dimensional Cartesian coordinate system in the northeast direction to obtain the state vector of the target ship. S2. Based on the state vector of the target ship, establish a linear observation model and use Kalman filtering to make predictions and obtain the predicted observation values ​​of the target ship. S3. Based on the actual radar observations and the predicted observations of the target ship at the current moment, calculate the observation residuals, and combine the covariance of the prediction error projected onto the observation space with the observation noise covariance matrix to obtain the innovative covariance matrix; calculate the Mahalanobis distance based on the observation residuals and the innovative covariance matrix, and construct an elliptical threshold region centered on the predicted observation position. S4. Determine the validity of observations based on Mahalanobis distance and the elliptical threshold region: If the Mahalanobis distance is less than or equal to the threshold of the elliptical threshold region, that is, the current radar observation falls within the threshold region, it is determined to be a valid observation, and Kalman filtering is used to update the state and obtain the updated trajectory of the target ship. If the Mahalanobis distance is greater than the threshold of the elliptical threshold region, that is, the current radar observation does not fall within the threshold region, it is judged as a mismatched observation, and a short-term missed detection processing method based on prediction is executed to obtain the predicted trajectory of the target ship. S5. Set a sliding time window. When the number of mismatched observations within the sliding time window reaches a preset threshold, occlusion determination is made by combining the geometric relationship between the ship's position, the predicted target position, and the static obstacle boundary. If the geometric occlusion condition is met, the target is determined to be in an occlusion state and enters the occlusion maintenance mode. By adjusting the filtering parameters and fusing auxiliary information when it is available, the target ship trajectory is continuously estimated to obtain the estimated trajectory of the target ship. Otherwise, continue with the short-term missed detection handling; S6. When several consecutive radar observations fall within the threshold area and the current geometric obstruction is removed, the obstruction is determined to have ended. The Kalman filter parameters are then gradually restored through a smooth transition, allowing the target trajectory to transition from the obstruction-maintained state to the normal tracking state.

[0006] Furthermore, the specific steps to obtain the predicted observations of the target ship include: S21. Based on the target ship's state vector, the state vector of the target ship satisfying the state transition matrix is ​​obtained through a two-dimensional planar uniform velocity motion model. The expression is:

[0007] In the formula, Let be the state vector of the target ship at the current moment; Let be the state vector of the target ship at the previous moment; This is the state transition matrix; This is process noise; S22. Based on the state vector of the target ship that satisfies the state transition matrix, construct a linear observation model, the expression of which is:

[0008] In the formula, This represents the radar observation vector at the current moment; The observation matrix; For radar observation noise, and ,in, The radar observation noise covariance matrix; S23, based on the posterior state estimate and posterior covariance matrix of the target ship at the previous moment, uses the state transition matrix and process noise covariance matrix through the Kalman filtering prediction step to calculate the prior state estimate and prior covariance matrix of the target at the current moment. The expression is as follows:

[0009]

[0010] In the formula, Estimate the prior state of the target at the current moment; Estimate the posterior state of the target at the previous time step; The prior covariance matrix; The posterior covariance matrix; The process noise covariance matrix; S24. Based on the prior state estimate of the target at the current moment, by replacing the target ship's state vector at the current moment in the linear observation model and setting the radar observation noise value to 0, the predicted observation value of the target ship is obtained, expressed as:

[0011] In the formula, These are the predicted observations for the target ship.

[0012] Furthermore, the specific steps of step S3 include: S31. Based on the radar observation vector at the current moment and the predicted observation value of the target ship, obtain the observation residual, expressed as:

[0013] In the formula, To observe residuals; S32. Based on the predicted observations of the target ship, combined with the covariance matrix after the prediction error is projected onto the observation space and the observation noise covariance matrix, the innovation covariance matrix is ​​obtained, expressed as:

[0014] In the formula, To innovate the covariance matrix; The covariance matrix is ​​the result of projecting the prediction error onto the observation space. S33. Based on the observation residuals and the innovation covariance matrix, calculate the Mahalanobis distance between the current observation position and the predicted position. The expression is:

[0015] In the formula, This represents the Mahalanobis distance between the current observation position and the predicted position. To create the inverse matrix of the covariance matrix; S34. Based on the Mahalanobis distance between the current observation position and the predicted position, set a threshold value; based on the predicted observation value of the target ship and the innovative covariance matrix, combined with the preset threshold value, construct an elliptical threshold region centered on the predicted observation position of the target ship, expressed as:

[0016] In the formula, This is the threshold value.

[0017] Furthermore, the steps for obtaining the estimated trajectory of the target ship are as follows: S41. Determine if the target ship has valid AIS information: If the target ship has valid AIS information, the position and velocity of the target ship in the valid AIS are used as observation vectors, and continuous estimation is performed through Kalman filtering to obtain the estimated trajectory of the target ship. If the target ship does not have valid AIS information, the filter parameters are adjusted. S42. Based on the process noise covariance matrix in Kalman filtering, and using the process noise amplification factor, the adjusted process noise covariance matrix is ​​obtained, expressed as:

[0018] In the formula, The process noise covariance matrix is ​​adjusted; This is the process noise amplification factor; S43. Based on the radar observation noise covariance matrix, and using the radar observation noise amplification factor, the adjusted radar observation noise covariance matrix is ​​obtained, expressed as:

[0019] In the formula, The adjusted radar observation noise covariance matrix; This represents the radar observation noise amplification factor. S44. Based on the prior state estimate of the target at the current moment, calculate the estimated observation vector through the linear observation model to obtain the estimated radar observation vector. S45. Based on the estimated radar observation vector, the adjusted process noise covariance matrix, and the adjusted observation noise covariance matrix, continuous estimation is performed using Kalman filtering to obtain the estimated trajectory of the target ship.

[0020] Furthermore, the method used to determine whether the geometric occlusion condition is met is as follows: S51. Represent the boundary line of an obstacle as a set of line segments; S52. Based on the current position of the ship and the predicted position of the target ship at the current time, construct the observation path between the ship and the target ship; S53. When the observation path between the ship and the target ship intersects with a set of line segments of the boundary line of the obstacle, the observation path is blocked, that is, the geometric occlusion condition is satisfied; otherwise, the geometric occlusion condition is not satisfied.

[0021] Beneficial effects: The present invention provides a stable radar target tracking method under obstructed environments. By establishing a two-dimensional plane rectangular coordinate system in the northeast direction and using Kalman filtering for prediction and updating, combined with effective observation and screening of elliptical threshold regions, it can effectively address the problems of discontinuity and mismatch of radar signals caused by obstruction of obstacles, and significantly improve the tracking continuity of targets in complex environments. A sliding time window is used to count the number of mismatched observations, and the occlusion is determined by combining the geometric relationship between the ship's position, the predicted target position, and the static obstacle boundary. This avoids misjudging short-term random missed detections as occlusions, improves the accuracy of occlusion detection, and reduces false alarms and erroneous state switching. For mismatched observations that have not fallen into the threshold area, a prediction-based short-term missed detection processing method is adopted. The predicted value of Kalman filtering is used to maintain the trajectory, which can effectively suppress trajectory interruption or jump caused by short-term signal loss and ensure the smoothness of tracking. Once the system determines that it has entered an obstruction state, it automatically switches to obstruction maintenance mode. By adjusting the filtering parameters and using the target predicted trajectory or auxiliary information, it continuously estimates the target ship's trajectory, enabling the system to reliably output the target status even when actual observation values ​​cannot be obtained, thus avoiding trajectory loss or divergence. When several consecutive frames of radar observation fall back into the threshold area and the geometric obstruction is lifted, the Kalman filter parameters are gradually restored through a smooth transition, allowing the target trajectory to smoothly transition from the obstruction maintenance state to the normal tracking state, avoiding trajectory jumps or filtering instability caused by parameter abrupt changes. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the radar target stabilization tracking method of the present invention; Figure 2 This is a diagram defining the coordinate system and target state in an embodiment of the present invention; Figure 3 This is a schematic diagram of elliptical gating based on predicted position in an embodiment of the present invention; Figure 4 This is a schematic diagram of target occlusion prediction based on geometric relationships in an embodiment of the present invention; Figure 5 This is a flowchart of the overall algorithm in an embodiment of the present invention; Figure 6 This is a schematic diagram of the obstacle drawing interface in an embodiment of the present invention; Figure 7 This is a schematic diagram of the visualization interface for the occlusion experiment results in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This embodiment provides a method for stable radar target tracking in obstructed environments, such as... Figure 1 As shown, it includes: S1. Establish a two-dimensional Cartesian coordinate system in the northeast direction, and map the motion state of the target ship observed by the radar to the two-dimensional Cartesian coordinate system in the northeast direction to obtain the state vector of the target ship. Specifically, such as Figure 2 As shown, a coordinate system is defined for the ship and the target ship to represent the relative position and motion parameters between the two ships; In this embodiment, a two-dimensional Cartesian coordinate system is established within the target tracking area, where the horizontal axis is the east direction coordinate axis and the vertical axis is the north direction coordinate axis; let the state vector of the ship at the current moment be:

[0026] In the formula, This is the state vector of the ship at the current moment; The coordinates are the ship's position due east. These are the coordinates of the ship's due north direction. This represents the velocity component of the ship in the due east direction; This represents the velocity component of the ship in the due north direction; Let the state vector of the target ship at the current moment be:

[0027] In the formula, Let be the state vector of the target ship at the current moment; The coordinates are the coordinates of the target ship's position due east. The coordinates are the north coordinates of the target ship. The velocity component in the direction due east of the target ship; This represents the velocity component of the target ship in the direction of due north.

[0028] S2. Based on the state vector of the target ship, establish a linear observation model and use Kalman filtering to make predictions and obtain the predicted observation values ​​of the target ship. Specifically, after establishing the coordinate system and defining the target state, it is necessary to further construct a target observation model and combine it with Kalman filtering to realize the recursive estimation of the target state. Its main function is to use the historical state of the target to predict the target position at the current moment, and then combine it with the current radar observation to determine whether the observation is reliable, and then decide whether to update the target state.

[0029] Preferably, the specific steps for obtaining the predicted observations of the target ship include: S21. Based on the target ship's state vector, the state vector of the target ship satisfying the state transition matrix is ​​obtained through a two-dimensional planar uniform velocity motion model. The expression is:

[0030] In the formula, Let be the state vector of the target ship at the current moment; Let be the state vector of the target ship at the previous moment; This is the state transition matrix; Process noise, used to represent the deviation between the actual motion of the target and the ideal model, such as disturbances caused by sea state, maneuvering behavior, or unmodeled maneuvering factors, can be represented as... ,in, The process noise covariance matrix; The state transition matrix can be represented as:

[0031] In the formula, The time interval between adjacent sampling times; Specifically, according to the two-dimensional planar uniform velocity motion model, within one sampling period, the position of the target ship is determined by the position and velocity of the previous moment, while the velocity remains unchanged within a short time interval. This can reflect the continuous motion characteristics of the ship target and is also convenient for subsequent integration with the Kalman filter recursive framework.

[0032] S22. In terms of observation, radar usually provides the target's position measurement at the current moment directly; Therefore, based on the state vector of the target ship that satisfies the state transition matrix, a linear observation model is constructed, expressed as:

[0033] In the formula, This represents the radar observation vector at the current moment; The observation matrix; For radar observation noise, and This indicates that the radar observation noise is modeled as zero-mean Gaussian noise, where, This is the radar observation noise covariance matrix, used to reflect the statistical characteristics of radar measurement errors; Since radar observation only involves the target's position and does not directly provide velocity information, the observation matrix can be written as: ; Accordingly, the radar observation vector at the current moment can be written as:

[0034] In the formula, This indicates the location of the target ship as observed by this vessel via radar. Specifically, since maritime radar observations are easily affected by factors such as sea clutter, noise interference, and target echo fluctuations, the actual observed values ​​often deviate from the true location. Therefore, it is necessary to use filtering prediction and threshold determination mechanisms to improve the reliability of observation utilization.

[0035] S23. At each sampling time, the target posterior state estimate and covariance matrix obtained at the previous time are first used to make a priori prediction of the state at the current time. Therefore, based on the posterior state estimate and posterior covariance matrix of the target ship at the previous moment, through the prediction step of Kalman filtering, using the state transition matrix and process noise covariance matrix, the prior state estimate and prior covariance matrix of the target at the current moment are calculated, as expressed in the following expression:

[0036]

[0037] In the formula, Estimate the prior state of the target at the current moment; Estimate the posterior state of the target at the previous time step; The prior covariance matrix; The posterior covariance matrix; The process noise covariance matrix; The above two equations respectively give the prediction results of the target state and state uncertainty without considering the new observation information at the current moment; through the prediction step of Kalman filtering, the most likely position of the target at the current moment and its error range can be obtained; S24. Based on the prior state estimate of the target at the current moment, by replacing the target ship's state vector at the current moment in the linear observation model and setting the radar observation noise value to 0, the predicted observation value of the target ship is obtained, expressed as:

[0038] In the formula, These are the predicted observations for the target ship.

[0039] Specifically, the predicted observations of the target ship correspond to Figure 2 The predicted position in the radar has the physical meaning of: the expected estimate of the target's current radar observation position based on historical tracking results and target motion model; the predicted position is not only the center of subsequent threshold determination, but also an important basis for occlusion prediction and trajectory continuity maintenance.

[0040] S3. Based on the actual radar observations and the predicted observations of the target ship at the current moment, calculate the observation residuals and combine them with the prediction covariance matrix and the observation noise covariance matrix to obtain the innovative covariance matrix; calculate the Mahalanobis distance based on the observation residuals and the innovative covariance matrix, and construct an elliptical threshold region centered on the predicted observation position. Preferably, step S3 includes the following specific steps: S31. Based on the radar observation vector at the current moment and the predicted observation value of the target ship, obtain the observation residual, expressed as:

[0041] In the formula, To observe residuals; Specifically, the observation residual reflects the magnitude of the deviation between the actual observation and the predicted observation; if the deviation is less than the threshold, it indicates that the current observation is consistent with the predicted trajectory and has high reliability; if the deviation is greater than the threshold, it indicates that the observation may have been affected by interference or clutter, or that the target state has changed abnormally. S32. Based on the predicted observations of the target ship, combined with the covariance matrix after the prediction error is projected onto the observation space and the observation noise covariance matrix, the innovation covariance matrix is ​​obtained, expressed as:

[0042] In the formula, To innovate the covariance matrix; The covariance matrix is ​​the result of projecting the prediction error onto the observation space. Specifically, the innovative covariance matrix comprehensively characterizes the overall uncertainty in the current observation space caused by both state prediction error and radar measurement error. That is, the innovative covariance matrix not only describes the degree of uncertainty of the predicted position, but also considers the noise influence of the radar measurement itself, and is the basis for constructing the statistical threshold region. S33. Based on the observation residuals and the innovation covariance matrix, calculate the Mahalanobis distance between the current observation position and the predicted position. The expression is:

[0043] In the formula, This represents the Mahalanobis distance between the current observation position and the predicted position. To create the inverse matrix of the covariance matrix; Specifically, the expression for calculating the Mahalanobis distance between the current observed position and the predicted position is a quadratic measure of the innovation vector after covariance normalization. Compared with ordinary Euclidean distance, Mahalanobis distance can simultaneously consider the distribution characteristics of observation errors in different directions, thus more accurately characterizing whether the current observation falls within a reasonable range of allowable errors. In target tracking problems, judging the validity of observations solely based on Euclidean distance often fails to reflect the fact that the ellipse extends to different degrees in different directions; while Mahalanobis distance, combined with the innovation covariance matrix, can make a more statistically meaningful judgment on the consistency between observations. S34. Based on the Mahalanobis distance between the current observation position and the predicted position, set a threshold value; based on the predicted observation value of the target ship and the innovative covariance matrix, combined with the preset threshold value, construct an elliptical threshold region centered on the predicted observation position of the target ship, expressed as:

[0044] In the formula, The threshold value is selected based on the two-dimensional chi-square distribution at a given confidence level.

[0045] Specifically, such as Figure 3As shown, the elliptical threshold region is a two-dimensional effective observation screening area composed of the predicted observation position as the center, the innovation covariance matrix as the shape template, and the Mahalanobis distance threshold as the radius parameter. Its shape is an elliptical region. Geometrically, the ellipse intuitively reflects the reasonable observation range near the target predicted position, and statistically, it represents a possible observation area at a given confidence level.

[0046] S4. Determine whether the current radar observation falls within the elliptical threshold region: If the current radar observation falls within the threshold area, it is determined to be a valid observation, and the Kalman filter is used to update the state and obtain the updated trajectory of the target ship. If the current radar observation does not fall within the threshold area, it is judged as a mismatched observation, and the predicted trajectory of the target ship is obtained through a short-term missed detection processing method based on prediction. The prediction-based short-term missed detection processing method discards current radar observations that fail to pass the elliptical threshold and outputs the target's prior state estimate at the current moment as the predicted trajectory of the target ship. At the same time, it records a missed event and increments the counter within the sliding window by 1. If the observations in the following frames return to normal, the counter is reset and normal tracking is resumed. If the number of missed events reaches the threshold, an occlusion determination is triggered.

[0047] In this embodiment, when the radar observation at the current moment satisfies When the condition is met, it indicates that the observation is within the elliptical threshold region and consistent with the current prediction state, so it can be determined as a valid observation and used for Kalman filter updates; when the condition is met... If the observation is outside the elliptical threshold region and does not match the predicted trajectory, the radar observation at the current moment may be clutter, false alarm, or abnormal echo. It may also mean that the target has experienced a deviation echo after a short-term missed detection, or has entered the obstacle-blocked area. Elliptical threshold determination is not only a means of eliminating abnormal observations, but also an important preliminary step in judging short-term missed detection and blockage status.

[0048] Specifically, for valid observations that pass the elliptic threshold test, the target state can be corrected according to the standard Kalman filter update formula; First, calculate the Kalman gain, expressed as:

[0049] In the formula, is the Kalman gain, which reflects the weight of the predicted value and the actual observation value in the current update; When the observation uncertainty is greater than the threshold and the observation noise is less than the threshold, the Kalman gain increases and the update result depends more on the current observation; conversely, when the observation noise is greater than the threshold or the observation confidence decreases, the Kalman gain decreases and the update result depends more on the predicted value. Subsequently, the prior state is corrected based on the current valid observations to obtain the posterior state estimate, expressed as: ; And update the corresponding covariance matrix, the expression is: ; Through the Kalman filter update process, the target state estimation is further optimized based on the fusion of predicted information and current observation information, thereby achieving smooth and stable tracking of the target position.

[0050] S5. Set a sliding time window. When the number of mismatched observations within the sliding time window reaches a preset threshold, occlusion determination is made by combining the geometric relationship between the ship's position, the predicted target position, and the static obstacle boundary. If the geometric occlusion condition is met, the target is determined to be in an occlusion state and enters the occlusion maintenance mode. The target ship's trajectory is continuously estimated by adjusting the filter parameters, the predicted trajectory of the target ship, or auxiliary information to obtain the estimated trajectory of the target ship. Otherwise, continue with the short-term missed detection handling; Specifically, such as Figure 4 As shown, occlusion determination based on geometric relationships is used to determine whether an occlusion state has been entered. After completing the elliptical threshold determination based on the predicted position, it is also necessary to combine the geometric relationship between the ship, the predicted position of the target, and the known obstacle boundary to determine whether the target has entered the occlusion state. The core purpose of this part is not to directly assume that the target is occluded based on the radar observation anomaly in a single frame, but to make a judgment by combining two conditions: continuous observation mismatch and the existence of occlusion relationship in the geometric relationship, so as to distinguish between short-term missed detection and real occlusion.

[0051] Preferably, the method used to determine whether the geometric occlusion condition is met is as follows: S41. Represent the boundary line of an obstacle as a set of line segments; S42. Based on the current position of the ship and the predicted position of the target ship at the current time, construct the observation path between the ship and the target ship; S43. When the observation path between the ship and the target ship intersects with a set of line segments of the boundary line of the obstacle, the observation path is blocked, which satisfies the geometric occlusion condition; otherwise, the geometric occlusion condition is not satisfied.

[0052] Specifically, geometric occlusion conditions alone are insufficient to indicate that a target has truly entered an occlusion state. In some cases, although the predicted target location is behind an obstacle, the current radar measurement may still be present temporarily. Conversely, even if the current measurement fails to pass the elliptical threshold, it may only be due to clutter interference or short-term missed detection. In this embodiment, the elliptical threshold determination results from the previous part are further combined for joint judgment. If the number of frames that fail to pass the threshold reaches a preset threshold within a set sliding window, and the geometric occlusion conditions are met at the same time, then the system determines that the target has entered an occlusion state. Therefore, the occlusion determination is not determined by a single factor, but by the following two conditions: First, the target has shown obvious observation mismatch in the most recent consecutive frames, that is, the current radar point continuously falls outside the predicted threshold ellipse; Second, the line of sight from the ship to the predicted position of the target is indeed blocked by a static obstacle in spatial geometry. Once the target vessel is confirmed to be in a state of obstruction, subsequent tracking can no longer rely entirely on the filtering parameters under normal conditions. The reason is that during the obstruction period, radar observation is usually unstable or even completely interrupted. If the parameter settings of the normal tracking phase are maintained, the prediction range may be too small, which may lead to the trajectory deviation accumulating too quickly, and it may even be difficult to recover smoothly when the target reappears. Therefore, adaptive adjustment is required during the obstruction phase.

[0053] The preferred method for obtaining the estimated trajectory of the target ship comprises the following steps: S51. Determine if the target ship has valid AIS information: If the target ship has valid AIS information, the position and velocity of the target ship in the valid AIS are used as observation vectors, and continuous estimation is performed through Kalman filtering to obtain the estimated trajectory of the target ship. If the target ship does not have valid AIS information, the filter parameters are adjusted. Specifically, when valid AIS information exists, AIS can be used as an auxiliary observation source to correct the target state during the obstruction phase; when radar observation is affected by obstruction, the position observation provided by AIS can be used first to participate in the state update; since AIS generally has high position accuracy, it can improve the trajectory continuity and position estimation stability during obstruction; if there is no AIS information, the system mainly relies on the prediction results after amplifying the noise in the process to maintain the continuity of the trajectory.

[0054] S52. Based on the process noise covariance matrix in Kalman filtering, and using the process noise amplification factor, the adjusted process noise covariance matrix is ​​obtained, expressed as:

[0055] In the formula, The process noise covariance matrix is ​​adjusted; This is the process noise amplification factor; Specifically, after entering the obstruction phase, the system appropriately increases the estimation of target uncertainty, making the prediction results have a wider range of coverage, thereby enhancing the ability to continuously maintain the trajectory under conditions of lack of reliable radar observation. Furthermore, when preparing to resume normal tracking from the obstruction state after the target reappears, the threshold also needs to be appropriately relaxed. Because when the target just reappears from behind the obstruction, there may still be some deviation between the actual radar observation point and the predicted trajectory. If the recovery conditions are set too strictly, it will result in a situation where the target has already appeared, but the system still cannot catch up. Therefore, a more relaxed threshold condition can be adopted in the recovery phase than in the normal phase, that is, a recovery threshold value greater than the preset value can be used. S53. Based on the radar observation noise covariance matrix, and using the radar observation noise amplification factor, the adjusted radar observation noise covariance matrix is ​​obtained, expressed as:

[0056] In the formula, The adjusted radar observation noise covariance matrix; This represents the radar observation noise amplification factor. The function of the adjusted radar observation noise covariance matrix is ​​to reduce the direct impact of a single radar observation on the filtering results in the early stage of recovery, making the process of the target trajectory returning from the occlusion maintenance state to the normal observation update state smoother, and avoiding trajectory abrupt changes due to the position jump of newly appearing radar points. S54. Based on the prior state estimate of the target at the current moment, calculate the estimated observation vector through the linear observation model to obtain the estimated radar observation vector. S55. Based on the estimated radar observation vector, the adjusted process noise covariance matrix, and the adjusted observation noise covariance matrix, continuous estimation is performed using Kalman filtering to obtain the estimated trajectory of the target ship.

[0057] S6. When several consecutive radar observations fall within the threshold area and the current geometric obstruction is removed, the obstruction is determined to have ended, and the Kalman filter parameters are gradually restored through a smooth transition, so that the target trajectory transitions from the obstruction maintenance state to the normal tracking state. Among them, a series of consecutive frames are preset positive integers, taking the value of 3 or 5; Specifically, if the target observation stably meets the recovery conditions within several consecutive frames, it is considered that the target has left the occlusion area, and the system enters the recovery phase or returns to the normal tracking state. To avoid significant trajectory jumps when the target recovers, the system incorporates a smooth transition process in the initial recovery phase. During this phase, the filter parameters do not immediately switch back from the obscured state to the normal state, but rather recover gradually, causing the weight of radar observations in the update to gradually increase from a low point until it returns to a normal level. The advantage of this approach is that it improves the smoothness of the trajectory during the recovery phase and reduces jumps and jitter.

[0058] Specifically, the overall flowchart of the radar target stabilization tracking method in this embodiment is as follows: Figure 5 As shown; First, multi-source input data is received, including radar observation data, AIS observation data, and ship motion information. Among them, radar observation is used to provide real-time position measurement of the target, AIS observation is used as an auxiliary high-precision information source for state correction when available, and ship motion information is used to establish the relative geometric relationship between the target and the ship, and to provide a basis for subsequent obstruction judgment. After the data input is completed, the MN-Kalman Filter is initialized, including the target initial state, covariance matrix, process noise parameters, observation noise parameters, and window length and threshold related to M / N judgment. At each sampling moment, the Kalman filter prediction step is first executed. Based on the target state estimate of the previous moment, the target state and prediction covariance at the current moment are predicted using the target motion model, and an elliptical threshold region is constructed near the predicted position. This elliptical threshold region is essentially determined by the prediction covariance and observation noise, and its function is to provide a reasonable observation range in which the target may appear at the current moment. Before entering the actual observation and discrimination, the system has already given a possible region in advance based on the target's historical motion trend, and all subsequent radar observations must first be compared with this region. In occlusion mode, if valid AIS information is available, the system further performs AIS-assisted updates. The introduction of AIS is not to replace radar, but to provide additional high-precision position reference for state estimation when radar observations are affected by occlusion and stability decreases. Since AIS usually has high position accuracy and continuity, introducing AIS observations during occlusion helps to suppress the cumulative error caused by relying solely on model predictions, and improves the stability and accuracy of the target trajectory during the occlusion phase. If AIS information is unavailable, the system can still continue to maintain continuous trajectory output by relying on the adjusted prediction model. After the target has been in the occlusion mode for a period of time, the subsequent radar observations are continuously monitored, and the occlusion end judgment and parameter recovery stage is entered. The function of this part is to determine whether the target has been freed from the obstacle occlusion and has re-entered a stable observation state; to detect whether the radar observations in the following several frames have re-stablely fallen into the relaxed threshold area, and at the same time to determine whether the current geometric occlusion conditions have been lifted; if the recovery conditions are met for several consecutive frames, the target has ended the occlusion and entered the recovery stage, and the filtering parameters are gradually and smoothly transitioned from the occlusion mode back to the normal tracking mode.

[0059] Specifically, the static obstacle drawing interface of the radar target stabilization tracking method in this embodiment is as follows: Figure 6 As shown, this is used to simulate obstacle occlusion under different conditions; After completing the target state modeling, ellipse threshold determination, and geometric occlusion recognition method design, in order to verify the radar target stabilization tracking method in a small static obstacle occlusion scenario, this embodiment further designed an obstacle drawing interactive interface for manually marking the boundaries of static obstacles within the target tracking area; by manually drawing the obstacle boundaries, a local occlusion scenario that meets the experimental requirements is constructed, thereby providing the necessary prior information of the scenario for subsequent geometric occlusion judgment; First, the ship's historical trajectory and the target's radar historical trajectory are displayed as background information in the same plane coordinate system, where the horizontal axis represents the position in the East direction and the vertical axis represents the position in the North direction. The light-colored scattered trajectory in the interface is used to show the approximate movement range of the ship and the target. Users can use this to determine where to place obstacles to create a local occlusion effect. Since there are no existing small obstacles in the original data, these background trajectories actually serve as a scene reference, allowing users to manually construct a reasonable static occlusion area by combining the target's movement path and the ship's line of sight.

[0060] Specifically, the visualization interface of the occlusion experiment results of the radar target stabilization tracking method in this embodiment is as follows: Figure 7 As shown, this interface is used to dynamically display and analyze the results of the entire tracking process. This interface corresponds to the comprehensive output window of the algorithm running stage. Its main function is to intuitively present the trajectory evolution process of the target in different states such as normal tracking, short-term missed detection, occlusion maintenance and recovery transition, so as to facilitate the observation of the continuous tracking effect of this method in small static obstacle occlusion scenarios. From the overall display content, the interface still uses the East-North Cartesian coordinate system, where the horizontal axis represents the position in the East direction and the vertical axis represents the position in the North direction. The interface simultaneously displays the obstacle position, the ship's trajectory, the real trajectory background, the radar observation point, the AIS observation point, the radar individually filtered trajectory, the AIS individually filtered trajectory, and the fused target trajectory, thus fully reflecting the algorithm's operating status at each stage.

[0061] Specifically, compared with traditional ship radar, the radar target stabilization tracking method in this embodiment has the following advantages: (1) It can effectively improve trajectory continuity in occluded scenes. Traditional radar target tracking methods typically rely heavily on valid measurements at the current moment. Once a target is obscured by a static obstacle in a local area, and stable observations cannot be obtained for several consecutive frames, problems such as trajectory interruption, restart, or target loss often occur. The radar target stable tracking method in this embodiment introduces an early occlusion determination and a continuous maintenance mechanism during occlusion based on target prediction. This allows the system to maintain trajectory continuity by relying on the predicted state and auxiliary information even when the target briefly enters behind an obstacle, thereby significantly improving the continuity of the target trajectory in small-scale occlusion scenarios. (2) It can effectively distinguish between short-term missed detections and real occlusions, improving the accuracy of judgment. In real-world maritime environments, radar observation anomalies do not necessarily originate from actual obstruction; they may also be caused by sea clutter, false alarms, echo fluctuations, or momentary missed detections. Traditional methods, which rely solely on single-frame anomalies or a small number of consecutive anomalies for target loss assessment, are prone to misjudgment. The radar target stabilization tracking method in this embodiment combines elliptical gating results with static obstacle geometric relationship analysis and utilizes M / N logic for continuous statistical judgment. This allows the radar target stabilization tracking method in this embodiment to not only focus on "whether the observation is abnormal," but also on "whether the anomaly persists" and "whether the anomaly has a reasonable obstruction scenario as a cause." Therefore, it can more accurately distinguish between ordinary short-term missed detections and actual obstruction, improving the reliability of obstruction identification. (3) It has the ability to detect occlusion in advance, thereby improving the system's proactive response capability. Traditional tracking methods mostly rely on passive compensation or re-acquisition after the target has been clearly lost, often resulting in delayed response and difficulty in recovery. The radar target stabilization tracking method in this embodiment utilizes the geometric relationship between the predicted target position and the static obstacle boundary to analyze whether the target is about to enter behind the obstacle before it is completely lost, thus achieving early detection of occlusion. This design transforms the system's handling of occlusion from "post-event compensation" to "early prediction and proactive response," enhancing the method's adaptability in complex scenarios. (4) Improve anomaly suppression capability through elliptical threshold regions The radar target stabilization tracking method in this embodiment constructs an elliptical threshold region near the predicted position to perform consistency checks on the current radar observations. Compared to directly using all radar measurements for updates, the radar target stabilization tracking method in this embodiment can prioritize filtering out clutter points, false alarm points, and abnormal observations that are significantly inconsistent with the target's predicted trajectory, thereby reducing the disturbance of unreliable measurements to the tracking results. This not only improves the stability during the normal tracking phase but also provides a unified gating basis for subsequent short-term missed detection processing and obstruction identification, enabling the radar target stabilization tracking method in this embodiment to maintain strong robustness even under conditions of fluctuating observation quality. (5) It has the ability to adaptively adjust parameters during occlusion, improving tracking stability in complex scenarios. When a target enters an obstructed state, using the same fixed filtering parameters as in the normal phase can easily lead to insufficient prediction range, state divergence, or difficulty in recovery. The radar target stabilization tracking method in this embodiment can adaptively adjust relevant parameters according to the characteristics of the current phase after confirming that the target has entered an obstructed state. This allows the system to appropriately relax its tolerance range for target motion uncertainty when stable radar observation is lacking, and to adopt a more robust observation access method in the recovery phase. Through this parameter adaptive mechanism, the trajectory maintenance capability during obstruction and the smoothness of the recovery phase can be better balanced, thereby improving the overall tracking stability. (6) Maintain stable trajectory output during the occlusion phase to reduce the difficulty of target re-acquisition. In traditional methods, once a target's trajectory is interrupted during the obstruction phase, subsequent reacquisition often requires re-association, restarting, or re-numbering, which increases tracking complexity and affects trajectory continuity. The radar target stabilization tracking method in this embodiment does not directly abandon the target during the obstruction phase, but continuously outputs the target trajectory through prediction maintenance, parameter adjustment, and necessary auxiliary observation support. In this way, even if the target is in a short-term unobservable state, the continuous evolution process of the target state is still preserved, reducing the difficulty of target reacquisition and re-association after the obstruction ends. (7) Improve overall tracking accuracy by using AIS-assisted fusion. The radar target stabilization tracking method in this embodiment is not limited to single radar observation. Instead, when AIS information is available, it is introduced into the target state estimation process as an auxiliary high-precision information source. In the normal phase, AIS can be used to assist in verifying the target trajectory. In the obstruction phase, AIS can provide additional reference constraints for the target position when radar observation is unstable or even missing. By using radar and AIS in fusion, the reliability of state estimation in obstruction scenarios is improved, and the error accumulation caused by relying solely on prediction propagation is reduced, thereby enhancing the overall accuracy and robustness of the system.

[0062] The present invention has the following beneficial effects: This invention provides a stable radar target tracking method under obstructed environments. By establishing a two-dimensional Cartesian coordinate system in the northeast direction and using Kalman filtering for prediction and updating, combined with effective observation filtering of elliptical threshold regions, it can effectively address the problems of discontinuity and mismatch of radar signals caused by obstruction of obstacles, and significantly improve the tracking continuity of targets in complex environments. A sliding time window is used to count the number of mismatched observations, and the occlusion is determined by combining the geometric relationship between the ship's position, the predicted target position, and the static obstacle boundary. This avoids misjudging short-term random missed detections as occlusions, improves the accuracy of occlusion detection, and reduces false alarms and erroneous state switching. For mismatched observations that have not fallen into the threshold area, a prediction-based short-term missed detection processing method is adopted. The predicted value of Kalman filtering is used to maintain the trajectory, which can effectively suppress trajectory interruption or jump caused by short-term signal loss and ensure the smoothness of tracking. Once the system determines that it has entered an obstruction state, it automatically switches to obstruction maintenance mode. By adjusting the filtering parameters and using the target predicted trajectory or auxiliary information, it continuously estimates the target ship's trajectory, enabling the system to reliably output the target status even when actual observation values ​​cannot be obtained, thus avoiding trajectory loss or divergence. When several consecutive frames of radar observation fall back into the threshold area and the geometric obstruction is lifted, the Kalman filter parameters are gradually restored through a smooth transition, allowing the target trajectory to smoothly transition from the obstruction maintenance state to the normal tracking state, avoiding trajectory jumps or filtering instability caused by parameter abrupt changes.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A radar target stable tracking method in a sheltered environment, characterized in that, include: S1. Establish a two-dimensional Cartesian coordinate system in the northeast direction, and map the motion state of the target ship observed by the radar to the two-dimensional Cartesian coordinate system in the northeast direction to obtain the state vector of the target ship. S2. Based on the state vector of the target ship, establish a linear observation model and use Kalman filtering to make predictions and obtain the predicted observation values ​​of the target ship. S3. Based on the actual radar observations and the predicted observations of the target ship at the current moment, calculate the observation residuals, and combine the covariance of the prediction error projected onto the observation space with the observation noise covariance matrix to obtain the innovative covariance matrix. Mahalanobis distance is calculated based on observation residuals and innovative covariance matrix, and an elliptical threshold region centered on the predicted observation location is constructed. S4. Determine the validity of observations based on Mahalanobis distance and the elliptical threshold region: If the Mahalanobis distance is less than or equal to the threshold of the elliptical threshold region, that is, the current radar observation falls within the threshold region, it is determined to be a valid observation, and Kalman filtering is used to update the state and obtain the updated trajectory of the target ship. If the Mahalanobis distance is greater than the threshold of the elliptical threshold region, that is, the current radar observation does not fall within the threshold region, it is judged as a mismatched observation, and a short-term missed detection processing method based on prediction is executed to obtain the predicted trajectory of the target ship. S5. Set a sliding time window. When the number of mismatched observations within the sliding time window reaches a preset threshold, occlusion determination is made by combining the geometric relationship between the ship's position, the predicted target position, and the static obstacle boundary. If the geometric occlusion condition is met, the target is determined to be in an occlusion state and enters the occlusion maintenance mode. By adjusting the filtering parameters and fusing auxiliary information when it is available, the target ship trajectory is continuously estimated to obtain the estimated trajectory of the target ship. Otherwise, continue with the short-term missed detection handling; S6. When several consecutive radar observations fall within the threshold area and the current geometric obstruction is removed, the obstruction is determined to have ended. The Kalman filter parameters are then gradually restored through a smooth transition, allowing the target trajectory to transition from the obstruction-maintained state to the normal tracking state.

2. The radar target stable tracking method in occluded environment according to claim 1, characterized in that, The specific steps to obtain the predicted observations of the target ship include: S21. Based on the target ship's state vector, the state vector of the target ship satisfying the state transition matrix is ​​obtained through a two-dimensional planar uniform velocity motion model. The expression is: wherein is the state vector of the target ship at the current time instant; is the state vector of the target ship at the previous time instant; is the state transition matrix; is the process noise; S22. Based on the state vector of the target ship that satisfies the state transition matrix, construct a linear observation model, the expression of which is: wherein is the radar observation vector at the current time instant; is the observation matrix; is the radar observation noise, and wherein is the radar observation noise covariance matrix; S23. Based on the posterior state estimate and posterior covariance matrix of the target ship at the previous moment, through the Kalman filtering prediction step, using the state transition matrix and process noise covariance matrix, calculate the prior state estimate and prior covariance matrix of the target at the current moment, expressed as: wherein is a prior state estimate of the target at the current time instant; is a posterior state estimate of the target at the previous time instant; is a prior covariance matrix; is a posterior covariance matrix; is a process noise covariance matrix; S24. Based on the prior state estimate of the target at the current moment, by replacing the target ship's state vector at the current moment in the linear observation model and setting the radar observation noise value to 0, the predicted observation value of the target ship is obtained, expressed as: In the formula, is the predicted observation of the target ship.

3. The radar target stable tracking method in occluded environment according to claim 2, characterized in that, The specific steps of step S3 include: S31. Based on the radar observation vector at the current moment and the predicted observation value of the target ship, obtain the observation residual, expressed as: In the formula, is the observation residual; S32. Based on the predicted observations of the target ship, combined with the covariance matrix after the prediction error is projected onto the observation space and the observation noise covariance matrix, the innovation covariance matrix is ​​obtained, expressed as: wherein is the innovation covariance matrix; is the covariance matrix of the prediction error projected into the observation space; S33. Based on the observation residuals and the innovation covariance matrix, calculate the Mahalanobis distance between the current observation position and the predicted position. The expression is: wherein is the Mahalanobis distance of the current observation position relative to the predicted position; is the inverse of the innovation covariance matrix; S34. Based on the Mahalanobis distance between the current observation position and the predicted position, set a threshold value; based on the predicted observation value of the target ship and the innovative covariance matrix, combined with the preset threshold value, construct an elliptical threshold region centered on the predicted observation position of the target ship, expressed as: In the formula, This is the threshold value.

4. The radar target stabilization tracking method under obstructed conditions according to claim 3, characterized in that, The steps to obtain the estimated trajectory of the target ship are as follows: S41. Determine if the target ship has valid AIS information: If the target ship has valid AIS information, the position and velocity of the target ship in the valid AIS are used as observation vectors, and continuous estimation is performed through Kalman filtering to obtain the estimated trajectory of the target ship. If the target ship does not have valid AIS information, the filter parameters are adjusted. S42. Based on the process noise covariance matrix in Kalman filtering, and using the process noise amplification factor, the adjusted process noise covariance matrix is ​​obtained, expressed as: In the formula, The process noise covariance matrix is ​​adjusted; This is the process noise amplification factor; S43. Based on the radar observation noise covariance matrix, and using the radar observation noise amplification factor, the adjusted radar observation noise covariance matrix is ​​obtained, expressed as: In the formula, The adjusted radar observation noise covariance matrix; This represents the radar observation noise amplification factor. S44. Based on the prior state estimate of the target at the current moment, calculate the estimated observation vector through the linear observation model to obtain the estimated radar observation vector. S45. Based on the estimated radar observation vector, the adjusted process noise covariance matrix, and the adjusted observation noise covariance matrix, continuous estimation is performed using Kalman filtering to obtain the estimated trajectory of the target ship.

5. The radar target stabilization tracking method under obstructed conditions according to claim 3, characterized in that, The method used to determine whether the geometric occlusion condition is met is as follows: S51. Represent the boundary line of an obstacle as a set of line segments; S52. Based on the current position of the ship and the predicted position of the target ship at the current time, construct the observation path between the ship and the target ship; S53. When the observation path between the ship and the target ship intersects with a set of line segments of the boundary line of the obstacle, the observation path is blocked, that is, the geometric occlusion condition is satisfied; otherwise, the geometric occlusion condition is not satisfied.

Images