A target information fusion method for full situation awareness

Abstract

The present invention discloses a target information fusion method for full situational awareness. This method uses a laser radar to cover the observation blind spots of a navigation radar, fuses the target information detected by the laser radar and the navigation radar, and uses this fused target information to provide full situational awareness of targets that could obstruct the ship's navigation, thereby safely avoiding collisions. By adding a laser radar to detect close-range targets and fusing this information with the target information from the existing navigation radar, the present invention improves the ship's full situational awareness of targets throughout its voyage, safeguarding safe close-range collision avoidance and enhancing the ship's autonomous navigation capabilities.

Description

Technical Field

[0001] The present invention belongs to the field of marine navigation, and in particular relates to a target information fusion method oriented to full situational awareness. Background Art

[0002] Due to pulse transmission, traditional navigation radars have a close-range radar observation blind area (RadarObservationBlindArea / Zone). This refers to the area within the radar's minimum detection range. Within this range, no matter how large the target is, it cannot be detected. The radar observation blind area typically refers to the area at the lower edge of the radar antenna's radiation angle, where radar waves cannot reach. The size of this area depends primarily on radar performance, such as pulse width, and the height of the radar antenna. The maximum range of a navigation radar depends primarily on the radar antenna height, target size and shape, and reflective antennas.

[0003] Currently, the method of raising the radar antenna higher is often used to reduce the radar observation blind spot. However, this method cannot fundamentally solve the blind spot problem and cannot achieve effective full-process precise navigation. In particular, it cannot play its due role in effectively avoiding collisions with ships at close range, and there are major safety hazards. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the above-mentioned prior art by adding a laser radar to achieve close-range target detection and integrating it with the target information of the original navigation radar, thereby improving the ship's full-situation awareness of targets throughout the entire navigation process, providing a guarantee for the ship's close-range safe collision avoidance, and improving the ship's autonomous navigation capability.

[0005] To achieve the above-mentioned purpose of the invention, the present invention provides a target information fusion method for full situational awareness, which uses a laser radar to cover the observation blind spot of the navigation radar, fuses the target information detected by the laser radar and the navigation radar respectively, and performs full situational awareness of the targets that hinder the navigation of the ship based on the fused target information to safely avoid collisions.

[0006] Furthermore, the following steps are included: (1) spatiotemporal alignment: placing the navigation radar and the lidar in the same coordinate system and working simultaneously; (2) track association: associating the tracks of the same target detected by the lidar and the navigation radar into the same track; (3) track fusion: judging whether to fuse based on the target distance and the lidar detection distance. If fusion is required, the local track detected in real time is fused into the current system track.

[0007] Furthermore, in step (1), the consistent reference point of the navigation radar is set at the installation position of the laser radar.

[0008] Furthermore, in step (2), the target tracks detected by the laser radar and the navigation radar are compared, a credibility evaluation objective function is constructed, the credibility of the track is determined by taking the maximum error likelihood estimation, and the tracks within the confidence interval of the function are associated.

[0009] Furthermore, the credibility evaluation objective function is where dij is defined as the distance between tracks, CikCjk is the distance between the kth point of the i-th and j-th tracks, σ is the systematic error of radar detection, and F is the accuracy range at the kth point.

[0010] Furthermore, the maximum allowable error range is obtained by constructing a system error model of the radar, and a comparison is made to determine whether the track to be fused is within the maximum allowable error range.

[0011] Furthermore, the system error model is σ(X, Y), where

[0012]

[0013]

[0014] Where, ρ is the target true value; Δρ is the system polar coordinate error; ε ρ is the system random error; Δx, Δy are the system displacement errors.

[0015] Furthermore, in step (3), when the estimation errors between the tracks detected by the laser radar and the navigation radar are uncorrelated, the track fusion is performed using a correlation method or a Kalman filter method.

[0016] Furthermore, in step (3), when the estimation errors between the tracks detected by the laser radar and the navigation radar are correlated, an adaptive weighted averaging method is used to perform track fusion.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] The data collected by detection equipment such as navigation radar and lidar are fused into radar target information. The lidar completes close-range target detection, solving the close-range blind spot problem caused by pulse emission of traditional navigation radar, improving the ship's target full situational awareness capability, and providing the necessary information support and safety guarantee for the ship to effectively perform collision avoidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flowchart of spatiotemporal alignment according to an embodiment of the present invention;

[0020] Figure 2 Schematic diagram showing a comparison of the tracks of the same target detected by a laser radar and a navigation radar, respectively, in one embodiment of the present invention;

[0021] Figure 3 A schematic diagram of a speed / heading difference fitting curve in one embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of laser radar target attenuation in one embodiment of the present invention;

[0023] Figure 5 This is a schematic diagram of track fusion in one embodiment of the present invention;

[0024] Figure 6 The flowchart of one embodiment of the present invention is shown in FIG. DETAILED DESCRIPTION

[0025] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0026] 1. Time and space alignment

[0027] Radar targets are measured in a local polar coordinate system. Detection data and target information from multiple radars should be observed and integrated within a unified coordinate system. This unified coordinate system is the fundamental model for radar information fusion. Currently, mainstream lidars have a detection range of 100-150 meters and utilize the same polar coordinate system for data processing as navigation radars. This allows the two radars to be roughly considered to operate in the same plane. Therefore, the choice of spatial coordinate system has minimal impact on radar detection and target tracking.

[0028] Traditional magnetron navigation radars and lidars are mechanical scanning radars, and target detection is periodic. Navigation radar detection cycles are typically 2-2.5 seconds, while lidar rotation cycles are 0.1 seconds. For moving targets, the motion state and parameters vary depending on the time of detection. Therefore, target fusion requires aligning target state information to the same moment.

[0029] When a ship is sailing at high speed, its position may have shifted significantly during the navigation radar detection cycle. Therefore, when performing track fusion or collision avoidance calculation, it is necessary to perform track deduction based on the motion state.

[0030] like Figure 1 As shown, the spatiotemporal alignment can be done as follows:

[0031] (1) System CCRP (consistency common reference point) calibration: Since the navigation radar antenna sensor and the lidar sensor are installed at different locations, it is necessary to calibrate the reference points to the same point in space through X and Y axis calibration methods;

[0032] (2) Time calibration of each device: The navigation radar and lidar are time-aligned to ensure that the scan lines at the same moment are aligned in time and space;

[0033] (3) Motion state calculation: During the movement of the ship, the navigation radar and lidar sensors move with the ship. The detected motion state needs to be calculated before it can be consistent with the ground static reference object;

[0034] (4) Track space-time alignment: After the space, time, and motion states are solved, the space-time alignment tends to be consistent, providing a basis for associating the ship's track;

[0035] (5) Track correlation: Track correlation is performed on the information detected by different radars to ensure that the track is based on a stationary reference object on the ground.

[0036] 2. Track association

[0037] The target track information reported by the navigation radar and lidar may be duplicated. If not fused, it may affect autonomous collision avoidance and path planning. To address the special circumstances where the lidar detection range is small and the target is within the critical detection range, the echo intensity of a single detection target is weak and the reliability is low, a target track correlation method is used to fuse the target. Through long-term coherent data accumulation, continuous and correlated target dynamic data is obtained, improving the credibility of target fusion.

[0038] Track correlation combines target tracks reported by multiple radars into a single target track. For the same target, multiple radars should produce relatively consistent track information. Track delays, jumps, and losses should not affect the overall track quality and thus the target fusion solution.

[0039] like Figure 2 As shown in the figure, the two track lines tend to be consistent. One of them represents the target track detected by the navigation radar, and the other represents the track of the same target detected by the lidar. The detection contents of the two radars are basically the same, and the target motion trajectory can be fused.

[0040] Track combination is mainly achieved by comparing and judging the position, speed and heading of the target detected by two radars, constructing a credibility objective function, setting confidence intervals, fuzzy intervals and rejection intervals to give corresponding results.

[0041] The credibility of track positions is primarily determined using maximum error likelihood estimation. By constructing a radar system error model, the maximum allowable error range is determined. This is then compared to determine whether the track to be fused is within this allowable error range. Radar errors are primarily determined by factors such as measurement system error, positioning system error, orientation system error, and information transmission delay error. Radar measurement system error is a systematic error that varies with the radar's operating environment, meteorological conditions, and other factors. Positioning and orientation system errors are caused by transmission errors between the positioning and orientation sensors. Information transmission delay error includes errors caused by delays in electromagnetic wave transmission and reception, signal processing, and information transmission delay within the network.

[0042] To perform track association, we must first construct the position credibility evaluation objective function of track association:

[0043]

[0044] Where, d ij Defined as the distance between tracks, |C ik C jk | is the distance between the kth point of the i-th and j-th tracks. σ is the systematic error of radar detection, and F is the accuracy range at the kth point.

[0045] The main difficulty in evaluating the reliability of a position lies in constructing a system error model σ(X, Y). The error model is constructed by integrating the measurement system error, positioning system error, orientation system error, and information transmission delay error:

[0046]

[0047]

[0048] Where, ρ is the target true value; Δρ is the system polar coordinate error; ε ρ is the system random error; Δx, Δy are the system displacement errors.

[0049] According to the relevant radar specifications, the radar azimuth and distance accuracy are 1°, 30m or 1% of the range (whichever is greater), respectively. The accuracy range at the kth track point can be calculated as

[0050] For targets traveling in the same direction but on opposite sides of the ship, the speed and heading must be integrated to distinguish whether they belong to the same target based on their motion characteristics. The speed and heading differences of the target's tracks are calculated, and an approximate curve is obtained by minimizing the variance. For the speed and heading of the same target, the difference approximate curve should gradually stabilize and converge to within the allowable error range.

[0051] Speed ​​and heading estimation are mainly carried out using the least square method of difference, and the speed and heading difference v is calculated. ik -v jk , and obtain a set of speed differences (v1,c1),…,(v n ,c n ).

[0052] Set the convergence curve of the speed difference to v k =a·t k +b, satisfying:

[0053]

[0054] To minimize M(a,b), we can calculate:

[0055]

[0056] The ideal fitting curve should make the velocity difference v ij Gradually converges to the error range and is in a stable state. Heading difference c ij The convergence calculation method is similar, such as Figure 3 shown.

[0057] Since the detection distance of the laser radar is short, its detection accuracy and reliability gradually decrease when the target gradually moves away from the ship. Therefore, when fusion of tracks, it is necessary to consider the reliability coefficient objective function M(l k ).like Figure 4 As shown, the principle of laser radar is that the closer it is, the clearer it is, and the farther it is, the less clear it is. Laser radar is mostly used to detect close-range targets.

[0058] If the fusion credibility of two tracks M(d ij ,v ij ,c ij ) meets the confidence interval requirements, then the association can be performed. The association weight of each track is calculated based on M(d ij ,v ij ,c ij ) ij ,v ij ,c ij And the kth laser radar track credibility coefficient l k Sure.

[0059] 3. Track Fusion

[0060] As the target tracking area expands, tracking systems face numerous shortcomings and require optimization due to the limitations of a single sensor's tracking range and the potential for target loss. To meet these demands, multi-sensor collaborative tracking technology has begun to gain increasing attention and plays a more important role in the design of contemporary target tracking systems. At the same time, as the operating time of target tracking systems continues to increase, the external environment faced by the entire tracking process becomes more complex, such as changes in the moving target medium and obstacles between multiple targets. This can indirectly lead to a decrease in the performance of target tracking systems using homogeneous sensors, and may even lead to missed targets during target detection. To address this shortcoming, heterogeneous sensors should be used whenever possible during system design to detect and track the target's movement. Unlike single-sensor tracking, multi-sensor target tracking has significant advantages, the most notable of which is the ability to effectively utilize the redundancy and complementarity of measurement data from different sensors to enhance the tracking performance of the entire fusion system.

[0061] The purpose of fusion of navigation radar and lidar target information is achieved by fusing local track with system track.

[0062] like Figure 5 As shown in the figure, the data detected by the laser radar and the data detected by the navigation radar are fused. The two radars detect the same target, but the detection methods are different. The comprehensive track of the system is calculated through the formula. The fusion characteristics are:

[0063] (1) As long as the fusion center receives new local track data, the fusion algorithm extrapolates the state of the system track at the previous time point to the time point of receiving the local track;

[0064] (2) Then it is associated and fused with the newly received local track data to obtain the state estimate of the system track at the current time point, and form a new system track;

[0065] (3) When another set of local tracks is obtained, repeat the above steps;

[0066] (4) The problem of correlated estimation errors must be faced.

[0067] Correlated errors exist because they are associated with the track information of the two radar sensors. In real-world applications, any errors in the system track, such as those caused by the correlation or fusion of historical tracks, will reduce fusion accuracy. Decorrelation algorithms are necessary to remove errors caused by these operations.

[0068] The main forms of this error include:

[0069] (1) When the estimated errors between the two tracks being fused are uncorrelated, fusion becomes relatively simple. Each track data is treated as a measurement data with independent errors and fused with the other track data. Standard methods such as correlation and Kalman filtering can be used to perform track fusion operations.

[0070] (2) When there is a correlation between the estimation errors.

[0071] An adaptive weighted average track fusion algorithm is employed. This algorithm utilizes a distributed tracking architecture, combined with the ship's navigation platform, to ensure independent observations from the navigation radar and lidar. Therefore, the key to using a weighted track fusion algorithm is determining the weights, which directly impacts the performance of the resulting track fusion algorithm.

[0072] A commonly used method is equal-weighted fusion, which adds the equally weighted measurements from multiple sensors and then averages them to form the final fused value. This method can analyze the track data transmitted by sensors in real time. However, the selection of weights is inherently subjective, and performance may not be optimal when used in a display. This is because the value of a particular sensor may deviate significantly from the true value, causing the fusion result to deviate significantly from the true value.

[0073] The weighted track fusion algorithm is a method in which multiple sensors measure the parameters of the same target in a certain area, taking into account the local track data of all sensors, setting weights for each sensor according to certain criteria, and finally using weighted fusion to obtain the global optimal track estimate.

[0074] The weight calculation method is as follows:

[0075] Assume that there are n sensors measuring the parameters of the same target from different directions. The measurement values ​​of each sensor are Si (i = 1, 2, 3, ..., n), and the variances of each sensor are The measurements between sensors are independent of each other, that is, their observation errors of the same target are also independent. The local track estimate of the i-th sensor at the k-th time point is assumed is an unbiased estimate.

[0076] Assume that the weighting factor values ​​of each sensor are ω i (i=1,2,3…n). Therefore, the fusion algorithm is divided into two types according to whether the weights of each sensor are equal: one is the equal-weight fusion method, also known as the averaging method; the other is the unequal-weight fusion method.

[0077] Assume that the fused track estimate is Weight ω iSatisfies the following formula:

[0078]

[0079] When weights are equal:

[0080]

[0081] When the weights are not equal, the following analysis is required. The total mean square error after fusion is

[0082]

[0083] Because Si (i = 1, 2, 3..., n) are independent of each other and are unbiased estimates of s,

[0084] E[(ss p )(ss q )]=0 (p≠q, p=1,2,…,n; q=1,2,…,n)

[0085] therefore,

[0086]

[0087] From the above formula, we can get that the total mean square error is a multivariate quadratic function composed of various weighting factors, so the mean square error σ 2 It must have a minimum value. The solution of this minimum value is the extreme value evaluation of the multivariate function of the weighting factor.

[0088] According to the method of finding the extreme value of multivariate function, the relative weight factor when the total mean square error is minimized can be obtained as follows:

[0089]

[0090] The corresponding minimum mean square error is:

[0091]

[0092] The above analysis is based on the estimated track measurements of each sensor at a specific point in time. The optimal weight is related to the variance of each sensor, so the sensor variance needs to be calculated based on the track data input by the sensor.

[0093] Assume that the navigation radar sensor is p and the lidar sensor is q, and their corresponding measurement values ​​are s p , s q , the corresponding measurement error of the two sensors is α p , α q ,Right now

[0094] s p =s+v p sq =s+v q

[0095] Where α p , α q is zero-mean stationary noise. Then the variance of the navigation radar p is:

[0096]

[0097] Because v p , ν q are uncorrelated with each other, have a mean of 0, and are also uncorrelated with s, so s p , s q The mutual correlation coefficient R pq for:

[0098] R pq =E[s p s q ]=E[s 2 ]

[0099] The autocorrelation coefficient R of s pp satisfy:

[0100]

[0101] Subtracting the two equations gives

[0102]

[0103] Corresponding R pq 、R pp The solution of can be calculated from its time domain estimated value.

[0104] Assume that the number of targets measured by the two radars is k, R pq (k), R pp (k) represents the kth R pq 、R pp The value of

[0105]

[0106] Similarly,

[0107]

[0108] Therefore, R can be calculated based on the inference algorithm and the sensor measurement value. pq and R pp , thereby calculating the variance of the sensor and obtaining the optimal weight.

[0109] 4. Algorithm implementation approach

[0110] like Figure 6As shown, the information fusion of the two radars is realized through a software program to solve the problem of close-range blind spots caused by pulse emission of traditional navigation radar. The close-range target detection is completed by the lidar and integrated with the navigation radar target, thereby improving the ship's target full situational awareness capability and further ensuring the ship's autonomous navigation. Specifically, it includes the following steps: the first step is to set the navigation radar consistent reference point (CCRP) to the lidar installation position; the second step is to receive the target information transmitted by the navigation radar and the lidar; the third step is to determine whether the target distance is greater than 200 meters through the navigation radar target detection result. If it is greater than 200 meters, execute the fourth step, otherwise execute the fifth step; the fourth step is to upload the target distance and azimuth information detected by the navigation radar to the autonomous navigation system and re-execute the second step; the fifth step is to determine whether the target distance is less than 150 meters through the navigation radar target detection result. If it is less than 150 meters, execute the sixth step, otherwise execute the Seven steps; Step 6, upload the target distance and azimuth information detected by the lidar to the autonomous navigation system, and re-execute the second step; Step 7, traverse the targets within the range of 100 meters to 150 meters of the lidar, calculate the distance difference and azimuth difference, select the target with an azimuth difference less than 1 degree and calculate the distance difference, and select the target with the smallest distance difference; Step 8, determine whether the minimum distance difference is less than 25 meters. If so, upload the target distance and azimuth information at the ratio of the optimal weights of the navigation radar information and the lidar information. If not, output the target information detected by the navigation radar and the lidar at the same time, and re-execute the second step.

[0111] 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 it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A target information fusion method for full situational awareness, characterized in that: The laser radar covers the blind spot of the navigation radar, and the target information detected by the laser radar and the navigation radar are integrated. Based on the integrated target information, the full situational awareness of the target that is hindering the navigation of the own ship is carried out to avoid collision safely. The process includes the following steps: (1) Space-time alignment: Navigation radar and lidar are placed in the same coordinate system and work simultaneously; (2) Track association: Associating the tracks of the same target detected by the laser radar and navigation radar into the same track; (3) Track fusion: Determine whether to perform fusion based on the target distance and the laser radar detection distance. If fusion is required, the local track detected in real time will be integrated into the current system track. When the estimated errors between the tracks detected by the laser radar and the navigation radar are uncorrelated, the correlation method or Kalman filter method is used for track fusion; when the estimated errors between the tracks detected by the laser radar and the navigation radar are correlated, the adaptive weighted average method is used for track fusion.

2. The target information fusion method for full situational awareness according to claim 1 is characterized in that: In the step (1), the consistent reference point of the navigation radar is set at the installation position of the laser radar.

3. The target information fusion method for full situational awareness according to claim 1 is characterized in that: In the step (2), the target tracks detected by the laser radar and the navigation radar are compared, a credibility evaluation objective function is constructed, the credibility of the track is determined by using the maximum error likelihood estimation, and the tracks within the confidence interval of the function are associated.

4. The target information fusion method for full situational awareness according to claim 3 is characterized in that: The credibility evaluation objective function is: Where, d ij Defined as the distance between tracks, |C ik C jk | is the distance between the kth point of the i-th and j-th tracks, σ is the systematic error of radar detection, and F is the accuracy range at the kth point.

5. The target information fusion method for full situational awareness according to claim 4 is characterized in that: The maximum allowable error range is obtained by constructing a system error model of the radar, and then compared to determine whether the track to be fused is within the maximum allowable error range.

6. The target information fusion method for full situational awareness according to claim 5 is characterized in that: The system error model is σ(X, Y), where Where, ρ is the target true value; Δρ is the system polar coordinate error; ε ρ is the system random error; Δx, Δy are the system displacement errors.

Images