Point trace matching method based on relaxation algorithm

Abstract

The invention discloses a trace point matching method based on a relaxation algorithm, and the method can achieve the restoration and improvement of a multi-target overall trace point with a large number of false alarm and leak detection phenomena. The method comprises two parts of false alarm elimination and missing detection completion. In the false alarm rejection part, a sliding window method is firstly used, a plurality of continuous screen point sets to be subjected to trace point matching are included in a sliding window, then redundant trace points are removed by using an outlier rejection algorithm, then mutual trace point matching is performed by using an improved relaxation algorithm, and finally a result is updated on an original point set. In the case of missing detection for complementing all scores, the method comprises the following steps: firstly, performing vector construction on point sets of adjacent screens by using a group velocity, indicating the direction and distance of vectors, and complementing trace points if no measurement point is found at a corresponding position; the effectiveness of the method is verified through simulation, and the method can be used for solving the problems of false alarm missing detection and robust tracking of multi-target overall trace points.

Description

Technical Field

[0001] This invention relates to the field of radar multi-target tracking, specifically to a method for multi-target point matching and measurement repair. Background Technology

[0002] Multi-target tracking is a crucial step in the target detection process for acquiring information about multiple targets. Its main objective is to estimate the motion state of multiple moving targets using measurements. Multi-target tracking originated from single-target tracking, but with the deepening and expansion of its applications, single-target tracking can no longer meet the complex requirements of detection, and multi-target tracking has gradually replaced single-target tracking in radar detection. Multi-target tracking mostly employs data association algorithms, which can be divided into likelihood function algorithms and Bayesian statistical algorithms based on specific application methods. For example, a paper on branching algorithms for distinguishing and tracking multiple targets proposes an optimized data association algorithm that uses a likelihood function to achieve track bifurcation. Another paper on optimizing the tracking and data correlation performance of surveillance systems in dense multi-target environments proposes a nearest-neighbor data association algorithm that uses a Bayesian filter to calculate the correlation between measurements and targets. However, multi-target tracking environments are not without clutter and false alarms, which often result in false alarm measurements or missed target points in practical applications, greatly affecting the robustness and accuracy of target detection. Therefore, there is an urgent need for a point matching method for eliminating false alarms and correcting missed detections in multi-target detection, which can identify and correct false alarm points and missed detection points in target detection, so as to achieve robust detection and tracking of multiple targets. Summary of the Invention

[0003] To overcome the problems existing in the prior art, this invention provides a dot matching method based on relaxation algorithm, which can repair and improve the overall dot pattern of multi-targets with a large number of false alarms and missed detections.

[0004] This invention comprises two parts: false alarm rejection and missed detection completion. In the false alarm rejection part, the invention first uses a sliding window method to encompass multiple consecutive screen point sets to be matched within the sliding window. Next, an outlier removal algorithm is used to calculate the shortest distance from each point to its surrounding points and average these distances to obtain a threshold distance. This normalized shortest distance is used to remove outliers and fuse contiguous points. Then, a modified relaxation algorithm is used to select each point set within the sliding window as the master point set, and two other randomly selected point sets as matching point sets. The quality score of the point pairs between the master and matching point sets is calculated to perform mutual point matching. By excluding point pairs with high quality scores that are considered successfully matched, the location of false alarm points is found and removed. In the missed detection completion part, the invention uses group velocity to construct vectors for the point sets of adjacent screens. Each point in one screen's point set is used as the starting point, specifying the direction and distance of the vector. If no measurement point is found at the corresponding position in the next screen's point set, the point is completed.

[0005] The technical solution of this invention is: a dot matching method based on a relaxation algorithm, comprising:

[0006] Step 1: Add the newly input measurement points to the sliding window until the upper limit of the sliding window is reached, and perform point matching on the point set within the sliding window;

[0007] Step 2: Apply the outlier removal algorithm to the point set before and after point matching;

[0008] Step 3: Use the improved relaxation algorithm to perform point matching on the point set;

[0009] Step 4: Use the vector measurement completion algorithm to perform point trace matching on the point set.

[0010] Furthermore, in step one, the method for matching the point set within the sliding window includes:

[0011] (a) Sliding window matching: Assume the radar detects l frames of measurement point data {P1, P2, ..., P} l The upper limit of the sliding window is T, and the set of measurement points in the sliding window is {P}. s ,P s+1 ,...,P s+T-1}, then the self-point set P in the sliding window s Point set P s+T-1 All of them need to be used as the principal point set for a set of point matching.

[0012] A set of dot matching is defined as one set of points as the main set and two sets of points as matching sets. Each main set of points needs to be matched twice with the two matching sets.

[0013] (b) Point Set Selection: After determining the principal point set in the sliding window, the matching point set is selected. The selection method depends on the position of the principal point set in the sliding window. For example, selecting the point set P at position i+1 in the sliding window. s+i If the main point set is selected, then the matching point set P is the set of points located at positions Ti and T-i+1. s+T-i-1 and P s+T-i When i = 0, i.e., with P s When the main point set is P, the matching point set is P. s+T-1 Except for P in the sliding window s The set of any points outside the principal point set. The definitions of the principal point set and the matching point set are explained in the following sections.

[0014] Furthermore, in step two, an outlier removal algorithm is used to prune the point set before point matching and to organize the point set after point matching, including:

[0015] (a) Outlier removal: Let P = {p1, p2, ..., p...} mLet} be the set of points to be removed from, containing m points. Define p. i For the i-th point in the point set, for each point p i Find the shortest distance D from the point to the surrounding points. i ,Right now:

[0016] D i =min j≠i [d(i,j)] (1)

[0017] Where d(i,j) is p i to p j The distance between them.

[0018] For each point in the point set P, the shortest distance is calculated and then averaged to obtain the threshold distance T:

[0019]

[0020] This leads to the normalized distance C. i :

[0021] C i =D i / T (3)

[0022] Let C be the normalized distance of each point in the point set P. i Both are related to the fusion threshold C f With the rejection threshold C t For example, if C i <C f Then point p will be marked. i After comparing it with the nearest point, merge multiple points with the same mark; if C i >C t Then for point p i Remove them.

[0023] (b) Merging: When applying a merging mark to two points, it first checks if any of the points has already been marked. If not, a new merging mark is applied; if one point exists, the same merging mark as the previously marked point is applied to the remaining points; if both have been marked, no further merging is performed. When merging points, the center positions of multiple points are selected, which is the cumulative average of their positions on the x and y axes.

[0024] (c) Removal: Permanently remove the point from the point set.

[0025] Furthermore, in step three, the improved relaxation algorithm is used to perform point matching on the point set; the improvements include:

[0026] (a) Matching point patterns: Let there be a principal point set P = {p1, p2, ..., p...} m} and matching point set and Match the principal point set P with the matching point set Q1 to obtain the point set P consisting of unmatched points. m and Q 1m , set P m and Q 1m The set of points to be determined is called the undetermined set U. The undetermined set U is merged with the principal set P to form a new principal set P. N Then P N Match the dots with Q2, P N The points that are retained in the old and new versions are the final results of the comparison point matching.

[0027] (b) Improved relaxation algorithm:

[0028]

[0029] Where r is the number of iterations.

[0030] T s =Kmax[S [r] (p i ,q j (5)

[0031] Where K∈(0,1) is the scaling parameter, max[S [r] (p i ,q j [T] represents the maximum quality score among all point pairs matched between two point sets P and Q. s This is the quality score threshold.

[0032] If among all matched point pairs, max[S] [r] (p i ,q j )]>T s Then p is considered i With q j If a match is found, the matching is disqualified from being repeated. The remaining points are then matched repeatedly until max[S] is reached. [r] (p i ,q j )]<T s If the two sets of points are not matched, then it is considered that there are no points in these two sets that can be matched.

[0033] In the second point matching comparison, the calculation method for the quality score of points that have not yet been matched differs. The formula for calculating the quality score in this case is as follows:

[0034]

[0035] Where, k own ∈(0,1), k other ∈(0,1), k own To influence the self-weight value, k other To influence the weight value of individuals, P Ui Let be the trace of points in the set of points U to be determined.

[0036] Furthermore, in step four, a vector measurement completion algorithm is used to repair the point set. Vectors between adjacent frames are constructed using defined directions and distances to complete any missed points between adjacent frames. Under slight dynamic changes, the effect of acceleration is negligible, and the direction and distance of the vectors are specifically determined by the group velocity v during the point set's movement.

[0037]

[0038] Among them, A a It is the average group direction, and the group velocity v per frame. i The cumulative average, where N is the number of frames of input measurement data, v ix With v iy It is the group velocity v per frame. i Components in the x and y directions.

[0039]

[0040] Among them, D a It is the average group displacement distance, and the displacement distance D of the cluster center at each frame point. i The cumulative average is given by Δt, where Δt is the inter-frame time, which is the radar scanning period.

[0041] Beneficial effects:

[0042] 1. This invention can adapt to complex configurations, multi-objective overall structures, and slight structural dynamic changes;

[0043] 2. Compared with existing algorithms, this invention has advantages in anti-interference, and can maintain high accuracy in false alarm and missed detection repair in high interference scenarios;

[0044] 3. This invention can perform joint correction of point matching and measurement repair on multiple targets as a whole. Unlike the monotonous process of traditional algorithms, this algorithm can assist each other in the correction process, improving the repair accuracy while iteratively outputting the correction results. Attached Figure Description

[0045] Figure 1 The above is a flowchart of the multi-target point matching method of the present invention.

[0046] Figure 2 This is a schematic diagram of the sliding window matching process of the present invention;

[0047] Figure 3 The diagram below illustrates the outlier removal algorithm of this invention.

[0048] Figure 4 The diagram below illustrates the point matching process of the improved relaxation algorithm of this invention.

[0049] Figure 5 The diagram below illustrates the process of differentiating false alarms and missed detections using the improved relaxation algorithm of this invention.

[0050] Figure 6 The diagram below is a schematic diagram of the vector measurement completion algorithm of the present invention.

[0051] Figure 7 This is a schematic diagram comparing simulated multi-target measurement data with true values ​​in a specific embodiment of the present invention;

[0052] Figure 8 The image shows a comparison between the multi-target point matching and repair results and the true values ​​in a specific implementation example of the present invention. Detailed Implementation

[0053] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0054] This invention discloses a multi-target tracking method. A single sensor continuously observes multiple targets within the same cluster, obtaining a multi-target measurement set. The invention uses an outlier removal algorithm to organize the multi-target measurement distribution and employs a modified relaxation algorithm and a vector measurement completion algorithm to jointly correct false alarms and missed detections in the multi-target measurements, thus achieving multi-target tracking. The method flow is as follows: Figure 1 As shown, the specific steps include the following:

[0055] Step 1: Use a sliding window to perform point matching on the point set. Add newly input measurement point sets to the sliding window until the maximum number of point sets in the sliding window is reached. The process is as follows: Figure 2 As shown. Simultaneously initialize parameters such as sliding window capacity W and fusion threshold C. f With the rejection threshold C t The number of iterations for the quality score is r, the proportionality parameter is K, and the influence factor is k. own k other wait.

[0056] Step 2: Before and after point matching, apply the outlier removal algorithm to the point set to be matched:

[0057] Let P = {p1, p2, ..., p} m Let} be the set of points to be removed from, containing m points. Define p.i For the i-th point in the point set P, for each point p i Find the shortest distance D from the point to the surrounding points. i ,Right now:

[0058] D i =min j≠i [d(i,j)] (9)

[0059] Where d(i,j) is p i to p j The distance between them.

[0060] For each point in the point set P, the shortest distance is calculated and then averaged to obtain the threshold distance T:

[0061]

[0062] This leads to the normalized distance C. i :

[0063] C i =D i / T (11)

[0064] Among them, C i >0, if 0 < C i ≤1, prove that the point p i Points that are relatively close to their surroundings may be merged, making them unsuitable for removal. If C... i >1, prove that point p i Points that are relatively far from their surroundings may be excluded and are not suitable for merging. The merging and exclusion process is as follows: Figure 3 As shown. The concepts of fusion and elimination will be explained in detail in the big claim.

[0065] Let C be the normalized distance of each point in the point set P. i Both are related to the fusion threshold C f With the rejection threshold C t For example, if C i <C f Then point p will be marked. i After comparing it with the nearest point, merge multiple points with the same mark; if C i >C t Then for point p i Remove them.

[0066] Step 3: Use the improved relaxation algorithm to perform point matching on the point set:

[0067] Let P = {p1, p2, ..., p}m} and Q = {q1,q2,...,q n Let} be two point sets for point matching, and calculate point p. i and q j between mass fractions s [0] (P i Q j ):

[0068]

[0069] Where, δ ij (h,k) refers to the point p. i and q j In addition, p h With q k The relative distance between two points

[0070]

[0071] It is point p h With q k For point p i and q j The quality score re-value evaluation function

[0072] Point p h With q k There is also a corresponding quality fraction s. [0] (p h ,q k ),and Combining these, we can obtain an updated revalued evaluation function.

[0073]

[0074] Using the updated revalue evaluation function To participate in the evaluation of the quality score of the person, so point p i and q j between mass fractions s [0] (p i ,q j Iterative updates are possible.

[0075]

[0076] Where r is the number of iterations.

[0077] T s =Kmax[S [r] (p i ,q j (17)

[0078] Where K∈(0,1) is the scaling parameter, max[S [r] (p i ,q j [T] represents the maximum quality score among all point pairs matched between two point sets P and Q. s Let S be the quality fraction threshold. [r] (p i ,q j )>T s Then p is considered i With q j If they match, then they don't match; otherwise, they don't match. The general process for matching points between two sets of points is as follows: Figure 4 As shown.

[0079] By performing a set of comparative point matching, matching a main point set twice with two matching point sets, the locations of false alarms and missed detections can be better identified. The undetermined point set U generated from the first point matching is used to distinguish between matched and unmatched points. The general process for this distinction is as follows: Figure 5 As shown.

[0080] In the second point matching, the calculation method for the quality score of points that have not yet been matched is different. The formula for calculating the quality score in this case is as follows:

[0081]

[0082] Where, k own ∈(0,1), k other ∈(0,1), k own To influence the self-weight value, k other To influence the weight value of individuals, P Ui Let U be the point trace in the undetermined point set U. For the definitions of principal point set, matching point set, and undetermined point set, please refer to the patent application form.

[0083] Step 4: Use the vector measurement completion algorithm to perform point matching on the point set. Construct vectors between adjacent frames using determined directions and distances to complete the missed points between adjacent frames. The general process is as follows: Figure 6 As shown. Under slight dynamic changes, the effect of acceleration is negligible, and the direction and distance of the vector are specifically determined by the group velocity v during the motion of the point set:

[0084]

[0085] Among them, A a It is the average group direction, and the group velocity v per frame. i The cumulative average, where N is the number of frames of input measurement data, v ix With v iy It is the group velocity v per frame.i Components in the x and y directions.

[0086]

[0087] Among them, D a It is the average group displacement distance, and the displacement distance D of the cluster center at each frame point. i The cumulative average is given by Δt, where Δt is the inter-frame time, which is the radar scanning period.

[0088] Each frame of time points has a corresponding group velocity. These parameters are used to construct vectors between adjacent frames, specifying the direction and distance of the vectors. If a point in a certain point set is not found near the corresponding position in the point set of the next frame after being translated in the direction and distance specified by the vector, it is considered that the position needs to be filled in and a measurement point needs to be added.

[0089] The following example, based on a single radar detecting multiple targets, illustrates and analyzes the algorithm. Assume a radar continuously observes a group of 15 targets within its detection area at a 1-second sampling interval, collecting 15 frames of measurement data. Each frame contains 3 false alarms (False Alarms), occurring within the x-axis and y-axis ranges of 0–150 m. Each frame also contains 2 missed detections, with the detected points allowed a 2-meter deviation from the true value. The multiple targets are arranged in a V-shaped formation with two structural partitions, the positive x-axis direction defined as 0°, and the two lobes extending towards 135° and -135° respectively. The entire group moves along the positive x-axis at a speed of 9 m / s. The overall structure of the multiple targets remains unchanged throughout the 15 frames until the simulation ends. A schematic diagram of the accumulated 15 frames of measurement data detected by the radar is shown below. Figure 7 As shown. This invention was applied to perform point matching and measurement repair experiments, where the algorithm parameters were set as follows: sliding window capacity W = 7, fusion threshold C... f The threshold C is 0.5. t The value is 3, the number of iterations r is 8, the scaling parameter K is 0.7, and the influence value k is... own 0.5, k other The value is 0.8. The results of dot matching and measurement repair are as follows: Figure 8 As shown, the present invention can achieve high-precision false alarm and missed detection repair for multi-target systems with complex configurations and slight structural dynamic changes.

[0090] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0091] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the scope of the appended claims.

[0092] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dot matching method based on a relaxation algorithm, characterized in that, include: Step 1: Add the newly input measurement points to the sliding window until the upper limit of the sliding window is reached, and perform point matching on the point set within the sliding window; Step 2: Apply the outlier removal algorithm to the point set before and after point matching; Step 3: Use the improved relaxation algorithm to perform point matching on the point set; Step 4: Use the vector measurement completion algorithm to perform point trace matching on the point set.

2. The method as described in claim 1, characterized in that, In step one, the sliding window matching method is used to match the points in the point set, including: (a) Sliding window matching: Assume the radar detects l frames of measurement point data {P1, P2, ..., P} l The upper limit of the sliding window is T, and the set of measurement points in the sliding window is {P}. s ,P s+1 ,...,P s+T-1 }, then the self-point set P in the sliding window s Point set P s+T-1 All of them need to be used as the principal point set for a set of point matching; A set of dot matching is defined as one set of points as the main set of points and two sets of points as matching sets of points. Each set of points needs to be matched twice with the two sets of matching points of points. (b) Point set selection: After determining the principal point set in the sliding window, the matching point set is selected. The selection method depends on the position of the principal point set in the sliding window; for example, selecting the point set P at position i+1 in the sliding window. s+i If the main point set is selected, then the matching point set P is the set of points located at positions Ti and T-i+1. s+T-i-1 and P s+T-i When i = 0, i.e., with P s When the main point set is P, the matching point set is P. s+T-1 Except for P in the sliding window s Any set of points outside the principal point set; the definitions of the principal point set and the matching point set are explained in the following sections.

3. The method as described in claim 1, characterized in that, In the second step, an outlier removal algorithm is used to prune the point set before point matching and to organize the point set after point matching, including: (a) Outlier removal: Let P = {p1, p2, ..., p...} m Let} be the set of points to be removed from, containing m points. Define p. i For the i-th point in the point set, for each point p i Find the shortest distance D from the point to the surrounding points. i ,Right now: D i =my j≠i [d(i,j)] (1) Where d(i,j) is p i to p j The distance between them; For each point in the point set P, the shortest distance is calculated and then averaged to obtain the threshold distance T: This leads to the normalized distance C. i : C i =D i / T (3) Let C be the normalized distance of each point in the point set P. i Both are related to the fusion threshold C f With the rejection threshold C t As, if C i <C f Then point p will be marked. i After traversing the process, merge multiple points with the same mark with the nearest point; if C i >C t Then for point p i Remove them; (b) Merging: When applying a merging mark to two points, it will first check whether one of the points has been merging marked before; if not, a new merging mark will be applied; if one point exists, the same merging mark as the point that has been merging marked will be applied to the remaining points; if both have been merging, no further merging will be applied; when merging points, the center position of multiple points will be selected, that is, the cumulative average of the positions on the x and y axes. (c) Removal: Permanently remove the point from the point set.

4. The method as described in claim 1, characterized in that, In step three, the improved relaxation algorithm is used to perform point matching on the point set; the improvement includes: (a) Yes, point matching: Let there be a principal point set P = {p1, p2, ..., p...} m } and matching point set and Match the principal point set P with the matching point set Q1 to obtain the point set P consisting of unmatched points. m and Q 1m , set P m and Q 1m The set of points to be determined is called the undetermined set U. The undetermined set U is merged with the principal set P to form a new principal set P. N Then P N Match the dots with Q2, P N The points that are retained in the old and middle parts are the matching points, which are the final results of the point matching; (b) Improved relaxation algorithm: Where r is the number of iterations; T s =Kmax[S [r] (p i ,q j )] (5) Where K∈(0,1) is a parameter, max[S [r] (p i ,q j [T] represents the maximum quality score among all point pairs matched between two point sets P and Q. s This is the quality score threshold; If among all matched point pairs, max[S] [r] (p i ,q j )]>T s Then p is considered i With q j If a match is found, the matching is disqualified from being repeated. The remaining points are then matched repeatedly until max[S] is reached. [r] (p i ,q j )]<T s If no match is found between the two point sets, then it is considered that there are no points in these two point sets that can be matched. Yes, in the second point matching process, the quality score calculation method differs for points that haven't yet matched. The formula for calculating the quality score in this case is as follows: Where, k own ∈(0,1), k other ∈(0,1), k own To determine the weight value of the response, k other In response to the weighting of human rights, P Ui Let be the trace of points in the set of points U to be determined.

5. The method as described in claim 1, characterized in that, In step four, a vector measurement completion algorithm is used to repair the point set. Vectors between adjacent frames are constructed using defined directions and distances to complete any missed points between frames. Under slight dynamic changes, the impact of acceleration is negligible, and the direction and distance of the vectors are specifically determined by the group velocity v during the point set's motion. Among them, A a It is the average group direction, and the group velocity v per frame. i The cumulative average, where N is the number of frames of input measurement data, v ix With v iy It is the group velocity v per frame. i Components in the x and y axes; Among them, D a It is the average group displacement distance, and the displacement distance D of the cluster center at each frame point. i The cumulative average is given by Δt, where Δt is the inter-frame time, which is the radar scanning period.

Images