A method for extracting target measurement sequence in clutter and missed detection environment

Abstract

The present invention discloses a method for extracting target measurement sequences in clutter and missed detection environments. This method utilizes prior information about targets, sensors, sensor platforms, and historical target measurement sequences, and proposes universal constraint functions for distance, angle, and Doppler measurements for missed detection situations, thereby achieving target measurement sequence extraction in missed detection environments. Furthermore, an innovative track splitting management strategy is proposed to achieve target measurement sequence extraction in clutter environments, ultimately achieving the goal of extracting target measurement sequences in clutter and missed detection environments. This method successfully achieves target measurement sequence extraction in complex environments. The algorithm generates only a small number of false tracks, with an average extraction accuracy of 98.8% and an average track continuity of 94.1%. The algorithm consumes only 0.033 seconds per frame on average, demonstrating strong practical engineering applicability.

Description

Technical Field

[0001] The present invention belongs to the field of automatic target tracking and relates to a method for extracting measurement sequences belonging to the same target at different times from a sensor measurement set with uncertain sources, and specifically relates to a method for extracting target measurement sequences in a clutter and missed detection environment. Background Art

[0002] In actual tracking environments, sensor measurement sets often face the problems of clutter interference and missed target detection. Clutter generally comes from two sources: 1) the natural environment, such as terrain and weather, in the same space as the sensor; and 2) other targets of no interest. Missed detections are generally caused by obstructions such as mountains and birds, or by the limitations of the sensor's own detection capabilities, resulting in certain frames in the sensor measurement set not containing target measurements but containing a large amount of clutter. To extract a sequence of measurements belonging to the same target at different times from a sensor measurement set with uncertain sources, it is necessary to filter out the clutter in each frame, leaving only the target measurements. However, the number of clutter measurements is often much greater than the number of target measurements, and each frame of clutter contains neither patterns nor any useful information. Therefore, it is very difficult to resolve the ambiguity of the source of target measurements in cluttered and missed detection environments. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention proposes a target measurement sequence extraction method in a clutter and missed detection environment. By measuring two or more frames before and after and combining other prior information, a rule constraint function is established to filter out clutter, thereby solving the problem of ambiguity in target measurement sources in clutter and missed detection environments.

[0004] The sensor is mounted on a movable sensor platform. The sensor measurement is a three-dimensional nonlinear measurement. The historical measurement of the sensor up to the kth moment is defined as Z k :

[0005] Z k ={Z1,Z2,...,Z k}

[0006] Among them, Z k Indicates that the current frame of the sensor at time k contains the measurements of clutter and target:

[0007]

[0008] in, represents the measurement of the sensor's current frame clutter at time k, represents the target measurement of the sensor in the current frame at time k.

[0009] A method for extracting target measurement sequences in a clutter and missed detection environment specifically comprises the following steps:

[0010] Step 1: Create a rule constraint function

[0011] From the three dimensions of distance measurement, angle measurement, and Doppler measurement, a rule constraint function is established for both missed detection and no missed detection, which is used for filtering missed detection and clutter. Missed detection means that after the rule constraint threshold is applied, no measurement point is framed within the threshold, indicating that the sensor in the current frame has missed detection of the target, and the target measurement sequence state in the current frame is recorded as an empty set. Assume that as of the current k n At this moment, the number of missed detection frames is a frames. When a=0, it means that there is no missed detection at present; when a≤N, it means that up to k n-a-1 When a>N, the target measurement sequence is deleted. N is determined by the M / N logic gate track management.

[0012] s1.1, distance measurement constraint function

[0013] The distance measurement d of the target at time k k for:

[0014]

[0015] in, is the target position vector, is the position vector of the sensor.

[0016] K n Distance measurement at time Breaks down to:

[0017]

[0018] Among them, the first part k n Noise-free distance measurement at all times, part 2 k n The distance measurement noise at the moment. According to the 3σ criterion, Scaling it, we get:

[0019]

[0020] K n Noise-free distance measurement at all times Breaks down to:

[0021]

[0022] Among them, the first group of variables and It is k n-a-1 Time distance measurement The components of the distance on the x and y axes are known a priori:

[0023]

[0024] in, represents k n-a-1 The sine value of the target angle measured by the sensor at that moment.

[0025] The second set of variables and It is k n-a-1 Time to K n The components of the sensor's motion distance at each moment in the x and y axes are known a priori:

[0026]

[0027] in, Indicates that the sensor is in k n-a-1 Time to K n The speed of movement at each moment, Indicates that the sensor is in k n-a-1 Time to K n The motion direction angle at each moment, T represents the sensor sampling period.

[0028] The third set of variables and It is k n-a-1 Time to K n The components of the target's movement distance at each moment on the x and y axes are unknown information:

[0029]

[0030] in, Indicates that the target is in k n-a-1 Time to K n The average speed at a given moment, Indicates that the target is in k n-a-1 Time to K n The unknown information contains two independent variables, namely the target k n-a-1 Time to K n The modulus and direction angle of the motion vector at each moment, and the above three groups of variables are sorted and substituted into the final noise-free distance constraint function:

[0031]

[0032] The value ranges of the two independent variables are:

[0033]

[0034] Among them, v T，max is the maximum moving speed of the target, ωT，max is the target maximum motion steering angle. and Solving the objective function using discretization of the independent variables A maximum and minimum range can be obtained, namely:

[0035]

[0036] Finally, we can get k n The threshold value of the distance measurement at any moment is:

[0037]

[0038] s1.2, Angle measurement constraint function

[0039] The angle measurement Θ of the target at time k k for:

[0040]

[0041] Among them, x T,k and y T,k They represent the x-direction and y-direction positions of the target at time k, respectively. S,k and y S,k They represent the x- and y-direction positions of the sensor at time k, respectively.

[0042] K n Angle measurement at time Breaks down to:

[0043]

[0044] According to the 3σ criterion, Scaling it, we get:

[0045]

[0046] right To break it down:

[0047]

[0048] Substituting the independent variable information, we get the noise-free angle constraint function:

[0049]

[0050] Through and Solving the objective function using discretization of the independent variables A maximum and minimum range can be obtained, namely:

[0051]

[0052] Will be solved and Back-substitution, we can get k n Threshold value for moment angle measurement:

[0053]

[0054] s1.3. Doppler measurement constraint function

[0055] The Doppler measurement γ of the target at time k k Defined as:

[0056]

[0057] in, is the velocity vector of the target, is the velocity vector of the sensor.

[0058] K n Doppler measurement at time Breaks down to:

[0059]

[0060] According to the 3σ criterion, Scaling it, we get:

[0061]

[0062] right Decompose it and get the noiseless Doppler constraint function:

[0063]

[0064] in, Indicates sensor k n-a-1 Doppler measurement of time, and represent the sensor and target k respectively n-a-2 Time to K n-a-1 The acceleration of time, is known a priori, is unknown, there are:

[0065]

[0066] Solving the objective function A maximum and minimum range can be obtained, namely:

[0067]

[0068] Will be solved and Back-substitution, we can get k n The threshold value of Doppler measurement at any moment:

[0069]

[0070] Step 2: Initialize the target measurement sequence

[0071] Target measurement initialization requires initializing two quantities: the state of the target measurement sequence and the track quality.

[0072] s2.1. In a real-world environment, new targets may appear or old targets may disappear at any moment. Therefore, a target measurement sequence must be initialized for each measurement point not framed by the threshold. Points not framed by the threshold are also called free measurement points. At the initial moment, all measurement points are free measurement points.

[0073] The state of the target measurement sequence is obtained from a free measurement point in the current frame. The state x of the target measurement sequence at time k is k for:

[0074]

[0075] s2.2. Track quality is an indicator used to measure the quality of the generated target measurement sequence. Assume that the current time is time k2. The state update of the target measurement sequence Γ1 was completed at time k1, and k1 < k2. Extract the measurement at time k2 to update the target measurement sequence Γ1.

[0076] First, use the three-dimensional rule constraint function obtained in step 1 to threshold the k2 moment measurement, and calculate the threshold value of k2 moment:

[0077]

[0078] The predicted value at time k2 is obtained by calculating the three sets of threshold values ​​at time k2 for:

[0079]

[0080] The state of the target measurement sequence Γ1 after the update at time k2 is Using the polar coordinate distance absolute value formula, calculate the state of the target measurement sequence Γ1 at time k2 and the predicted value of the target measurement sequence Γ1 at time k2 The distance between

[0081]

[0082] Among them, the polar coordinates of the first point are P1 = (ρ1, Θ1), and the polar coordinates of the second point are P2 = (ρ2, Θ2).

[0083] The target measurement sequence Γ i The track quality AQ is defined as the sum of the distances between the state and the predicted value of each frame:

[0084]

[0085] Where l is less than or equal to the maximum number of frames in each target measurement sequence.

[0086] Step 3: Splitting the target measurement sequence

[0087] When performing a threshold selection, various scenarios may occur: ① Only measurements containing the target are selected; ② Only measurements containing clutter are selected; ③ Measurements containing both clutter and the target are selected; ④ No points are selected. The target measurement extraction process requires extracting as many target measurements as possible without changing the original points. Therefore, the following target measurement sequence splitting method is designed: When no measurement points are selected within the threshold, the target measurement sequence state for that frame is set to an empty set. When the number of measurement points selected within the threshold is not zero, a target measurement sequence is updated for each measurement point within the threshold.

[0088] Step 4: Target measurement sequence management

[0089] In complex environments with dense clutter and potential missed detections, target measurement sequence management is a crucial measure for confirming target measurement sequences as confirmed tracks (CTs). The high maneuverability of targets, sensor limitations, and the emergence of false tracks complicate this process. Here, we employ an M / N logic rule for track management. In the past M consecutive sensor sampling cycles, if the target measurement sequence is not in the empty set state—that is, the number of frames containing measurement values ​​selected by the threshold is not less than a set N—then the target measurement sequence is marked as a CT and retained. If the number of frames in the target measurement sequence that are not in the empty set state is less than the set N, the target measurement sequence is marked as a false track and deleted. Target measurement sequences shorter than M frames are marked as unknown tracks and retained. The specific values ​​of M and N are determined based on the clutter density and missed detection probability in the scene.

[0090] Preferably, the value range of M is [5, 10], and N=0.5M.

[0091] Step 5: Merge target measurement sequences

[0092] After the management in step 4, the following three types of tracks still exist: ① Tracks obtained by splitting the target measurement sequence in step 3; ② Tracks of different lengths but with inclusion relationships; ③ False clutter tracks that remain after M / N logic management. Therefore, further screening and merging are required through target measurement sequence merging:

[0093] s5.1. For the track obtained by splitting the target measurement sequence in step 3, calculate its track quality after splitting N frames and compare it with the track quality of the original target measurement sequence. Save the sequence with lower track quality, i.e., higher quality.

[0094] s5.2. Because a target's measurement point initially initiates a target measurement sequence, and another target measurement sequence is initiated in a frame near that target in an intermediate frame, or a clutter source closer to the target's measurement point in a frame initiates a target measurement sequence with the same subsequent track state as the target, two tracks of different lengths may appear, but they contain each other. When multiple targets move in space, it is impossible for them to have the same state in the same frame; otherwise, a collision would occur. Therefore, the number of overlapping frames between two target measurement sequences is calculated. If two or more frames have identical track states, the target measurement sequence with the fewer frames is deleted, retaining only the target measurement sequence with the longer frame number.

[0095] s5.3, when the environmental clutter density is too high, M / N logic management cannot delete some false tracks. Therefore, the track quality is used to judge, first calculate the average track quality of each target measurement sequence frame by frame, and when the average track quality is greater than the set threshold AQ max When the track is detected, it is judged as a false track and deleted.

[0096] After the target measurement sequences are merged, the remaining track is the target measurement sequence extracted in the clutter and missed detection environment.

[0097] The present invention has the following beneficial effects:

[0098] 1. Distance, angle and Doppler measurement rule constraints are proposed. The joint constraints realize the extraction of target measurement and solve the problem of target measurement sequence extraction in missed detection environment.

[0099] 2. The track splitting management strategy is applied to the target measurement sequence extraction, which improves the accuracy of target measurement sequence extraction in the presence of clutter.

[0100] 3. The definition of track quality is proposed, which not only provides a scoring index for the tracks split from the track, but also provides a new solution for eliminating false tracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 Flowchart of the target measurement sequence extraction method;

[0102] Figure 2 Schematic diagram of a simulation scenario in an embodiment;

[0103] Figure 3 The upper and lower limits of the distance measurement obtained in the embodiment are shown;

[0104] Figure 4 The upper and lower limits of angle measurement obtained in the embodiment;

[0105] Figure 5 The upper and lower limit diagrams of Doppler measurements obtained in the embodiment;

[0106] Figure 6 This is a schematic diagram of the measurement and extraction results in the embodiment;

[0107] Figure 7 This is a partial enlarged view of the measurement and extraction results in the embodiment;

[0108] Figure 8 CT quantity map obtained for the embodiment;

[0109] Figure 9 Figure 2 is a graph of the number of CTTs obtained for the embodiment; DETAILED DESCRIPTION

[0110] The present invention will be further explained below with reference to the accompanying drawings;

[0111] A method for extracting target measurement sequences in clutter and missed detection environments, such as Figure 1 As shown, at time k1, a target measurement sequence is initialized for each measurement point. Each target measurement sequence initialized at time k1 is used to perform three types of constrained joint thresholds on the measurement points at time k2. A target measurement sequence is split for each framed measurement point. The track quality metric is used to manage and merge the split target measurement sequences, ultimately obtaining a high-precision target measurement sequence at time k2. The algorithm is then iterated by back substitution, ultimately extracting the target measurement sequence in complex environments.

[0112] The simulation scenario of this embodiment is as follows Figure 2 As shown in Figure 2, in a two-dimensional Cartesian coordinate system, the target moves from the lower left to the upper right, and the sensor moves from the lower right toward the target. This method was used to simulate the target, and the results of 50 Monte Carlo experiments and 200 simulation frames were analyzed.

[0113] To verify the effectiveness of the rule constraint function, the difference between the upper and lower thresholds of the three types was calculated in a simple environment, and the success rates of the three thresholds were compared. The difference between the upper and lower thresholds was calculated by subtracting the lower limit of the three thresholds for each of the 200 simulation frames. If the difference between the upper and lower thresholds is large, not only will the target measurement point be framed in, but also a large amount of clutter, resulting in a large number of false tracks when managing the target measurement sequence. Therefore, the smaller the difference between the upper and lower thresholds and the higher the success rate of framing the target measurement point within the joint threshold, the more effective the rule constraint function.

[0114]

[0115] Table 1

[0116] As shown in Table 1, after the simulation test in a simple environment, the threshold success rate reached 99.5%. Only one frame out of 200 frames failed to frame the threshold. This is because the distance measurement noise of the target measurement generated by the 187th frame that failed to frame the threshold is 4σ. d , which exceeds the 3σ criterion when establishing the rule constraint function and is a low-probability event.

[0117] Figure 3 、 4 Figures 5 and 5 show the upper and lower limits of the distance, angle, and Doppler measurements of the new algorithm under a 200-frame simple environment simulation experiment, showing the comparison between the upper and lower limits obtained in this embodiment and the actual target measurement. It can be clearly seen that the target measurement is within the upper and lower limits of the new algorithm threshold, and the threshold range is very small.

[0118] Simulation experiments were carried out in a complex environment with a clutter density of 2e-6 and a missed detection probability of 0.8. Figure 6 、 7 The figure shows the effect of the target measurement extraction algorithm of this embodiment in a complex environment. It can be clearly seen that the new algorithm has a good effect in extracting measurement values ​​in a complex environment.

[0119] Figure 8 、 9 The figure shows the CT and confirmed true track (CTT) numbers obtained when the actual number of targets is 1. In this embodiment, M=8 and N=4 are set in the M / N logic rule method. Therefore, the target measurement sequence management starts only when the number of track frames reaches 8, which is correspondingly reflected in the CT figure and CTT. Figure 1 The first eight frames are all 0. As can be seen from the figure, the average CT number of this embodiment is basically stable at around 1, and the ratio of CT to CTT is around 1, indicating that this embodiment generates fewer false tracks through effective rule constraint functions and track management strategies.

[0120] Table 2 shows the average track continuity and extraction accuracy of this embodiment under 50 Monte Carlo simulations and 200 simulation frames. It can be clearly seen that the track continuity and extraction accuracy of this method are close to 100%, and the real-time performance is also good, which can be applied to actual projects.

[0121]

[0122] Table 2

[0123] In order to verify the effectiveness of this method in different complex environments, tests were carried out under different missed detection probabilities and different clutter densities.

[0124]

[0125] Table 3

[0126] As shown in Table 3, maintaining the clutter density at 2e-6 and varying the missed detection probability, extraction accuracy and track continuity remain high when the missed detection probability is 0.9 and 0.8. When the missed detection probability begins to drop to 0.7 and 0.6, both accuracy and track continuity decline rapidly, but the per-frame time remains stable at an average level.

[0127]

[0128] Table 4

[0129] As shown in Table 4, the probability of missed detection remains constant at 0.8 while varying the clutter density. As the clutter density increases, the extraction accuracy and track continuity remain unchanged, while the algorithm's frame time increases exponentially. This suggests that the new algorithm can achieve relatively good extraction results even at high clutter densities, but its real-time performance may deteriorate.

[0130] The above simulations demonstrate that this method can be applied to various complex environments and is particularly effective for extracting target measurements in these environments. For example, it can be used to identify target tracks from the CT output of a target tracking algorithm when the true value is unknown. The CT output of a target tracking algorithm often contains a large number of false tracks. By first converting the CT output of the target tracking algorithm from the state domain to the measurement domain, the identified target measurements are compared with the CT using target measurement extraction techniques to select the target CT and eliminate other false tracks. This method can also be used to improve the tracking accuracy of target tracking algorithms. By inputting the measurements obtained through target measurement extraction into the target tracking algorithm, the performance of the target tracking algorithm can be significantly enhanced. In practical applications, this method can be applied to unmanned driving technology to improve millimeter-wave radar target tracking accuracy and enhance the safety of unmanned driving technology. It can also be applied to missile strikes to improve the accuracy of missile strikes.

Claims

1. A target measurement sequence extraction method in a clutter and missed detection environment. The sensor is mounted on a movable sensor platform. The sensor measurement is a three-dimensional nonlinear measurement. The historical measurement of the sensor up to time k is defined as Z k : WITH k ={Z1,Z2,...,Z k } in, Z k Indicates that the current frame of the sensor at time k contains the measurements of clutter and target: in, represents the measurement of the sensor's current frame clutter at time k, Represents the target measurement of the sensor in the current frame at time k; it is characterized by comprising the following steps: Step 1: Create a rule constraint function Based on the rule constraint function, it is used for filtering missed detection and clutter. The missed detection means that after the rule constraint threshold is set, there is no measurement point within the threshold, indicating that the sensor in the current frame misses the target, and the target measurement sequence state in the current frame is recorded as an empty set. Assuming that the current k n At this moment, the number of missed detection frames is a frames. When a=0, it means that there is no missed detection at present; when a≤N, it means that up to k n-a-1 When a>N, the target measurement sequence of the current frame is deleted; N is determined by the M / N logic gate track management; s1.1, distance measurement constraint function The distance measurement d of the target at time k k for: in, is the target position vector, is the position vector of the sensor; K n Distance measurement at time Breaks down to: Among them, the first part k n Noise-free distance measurement at all times, part 2 k n The distance measurement noise at the moment; according to the 3σ criterion, Scaling gives: K n Noise-free distance measurement at all times Breaks down to: Among them, the first group of variables and It is k n-a-1 Time distance measurement The components of the distance on the x and y axes are known a priori: in, represents k n-a-1 The sine value of the target angle measured by the sensor at that moment; The second set of variables and It is k n-a-1 Time to K n The components of the sensor's motion distance at each moment in the x and y axes are known a priori: in, Indicates that the sensor is in k n-a-1 Time to K n The speed of movement at each moment, Indicates that the sensor is in k n-a-1 Time to K n The motion direction angle at each moment, T represents the sensor sampling period; The third set of variables and It is k n-a-1 Time to K n The components of the target's movement distance at each moment on the x and y axes are unknown information: in, Indicates that the target is in k n-a-1 Time to K n The average speed at a given moment, Indicates that the target is in k n-a-1 Time to K n The average steering angle at the moment; the unknown information contains two independent variables, namely the target k n-a-1 Time to K n The modulus and direction angle of the motion vector at each moment, and the above three groups of variables are sorted and substituted into the final noise-free distance constraint function: The value ranges of the two independent variables are: Among them, v T，max is the maximum moving speed of the target, ω T，max is the target maximum motion steering angle; and Solving the objective function using discretization of the independent variables Get a maximum and minimum interval: Get k n The threshold value of the distance measurement at any moment is: s1.2, Angle measurement constraint function The angle measurement Θ of the target at time k k for: Among them, x T,k and y T,k They represent the x-direction and y-direction positions of the target at time k, respectively. S,k and y S,k They represent the x- and y-direction positions of the sensor at time k respectively; K n Angle measurement at time Breaks down to: According to the 3σ criterion, Scaling it, we get: right To break it down: Substituting the independent variable information, we get the noise-free angle constraint function: Through and Solving the objective function using discretization of the independent variables A maximum and minimum range can be obtained, namely: Will be solved and Back-substitution, we get k n Threshold value for moment angle measurement: s1.

3. Doppler measurement constraint function The Doppler measurement γ of the target at time k k Defined as: in, is the velocity vector of the target, is the velocity vector of the sensor; K n Doppler measurement at time Breaks down to: According to the 3σ criterion, Scaling it, we get: right Decompose it and get the noiseless Doppler constraint function: in, Indicates sensor k n-a-1 Doppler measurement of time, and represent the sensor and target k respectively n-a-2 Time to K n-a-1 The acceleration of time, is known a priori, is unknown, there are: Solving the objective function Get a maximum and minimum interval: Will be solved and Back-substitution, we get k n The threshold value of Doppler measurement at any moment: Step 2: Initialize the target measurement sequence Initialize the state of the target measurement sequence and the track quality respectively; s2.

1. The state of the target measurement sequence is obtained from a free measurement point in the current frame. The free measurement point is all the measurement points at the initial moment or the measurement points framed by the threshold in each frame. The state x of the target measurement sequence at time k is k for: s2.

2. Assume that the current time is time k2. The state update of the target measurement sequence Γ1 has been completed at time k1, and k1 < k2. Extract the measurement at time k2 to update the target measurement sequence Γ1. First, use the three-dimensional rule constraint function obtained in step 1 to threshold the k2 moment measurement and calculate the threshold value at k2 moment: The predicted value at time k2 is obtained by calculating the three sets of threshold values ​​at time k2 for: The state of the target measurement sequence Γ1 after the update at time k2 is Using the polar coordinate distance absolute value formula, calculate the state of the target measurement sequence Γ1 at time k2 and the predicted value of the target measurement sequence Γ1 at time k2 The distance between The target measurement sequence Γ i The track quality AQ is defined as the sum of the distances between the state and the predicted value of each frame: Where, l is less than or equal to the maximum number of frames of each target measurement sequence; Step 3: Splitting the target measurement sequence Use the rule constraint function obtained in step 1 to perform threshold selection; when no measurement point is selected within the threshold, set the target measurement sequence state of the frame to an empty set; when the number of measurement points selected within the threshold is not 0, update a target measurement sequence for each measurement point within the threshold; Step 4: Target measurement sequence management Select M / N logic rules to manage target measurement sequences and eliminate false tracks with too many missed frames; Step 5: Merge target measurement sequences All tracks are classified and merged according to the following steps: s5.

1. For the track obtained by splitting the target measurement sequence in step 3, calculate its track quality after splitting N frames and compare it with the track quality of the original target measurement sequence. Save the sequence with the lower track quality, i.e., the higher quality sequence; s5.

2. For tracks of different lengths but containing each other, calculate the number of overlapping frames between different target measurement sequences. If there are two or more frames with exactly the same track status, delete the target measurement sequence with fewer frames. s5.

3. For the track managed in step 4, first calculate the average track quality of each target measurement sequence by frame. When the average track quality is greater than the set threshold AQ max When the track is detected, it is judged as a false track and deleted; After the target measurement sequences are merged, the remaining track is the target measurement sequence extracted in the clutter and missed detection environment.

2. The method for extracting target measurement sequences in a clutter and missed detection environment as claimed in claim 1, wherein: The M / N logic rule is as follows: in the past M consecutive sensor sampling cycles, if the number of frames in the target measurement sequence that are not in the empty set state is not less than the set value N, then the target measurement sequence is marked as a target confirmation track and retained; If the number of frames in the target measurement sequence that are not in the empty set state is less than the set value N, the target measurement sequence is marked as a false track and deleted. For target measurement sequences with a length of less than M frames, they are marked as unknown tracks and retained. The specific values ​​of M and N are determined by the clutter density and missed detection probability in the scene.

3. The method for extracting target measurement sequences in a clutter and missed detection environment according to claim 1 or 2, wherein: In the M / N logic rule, the value range of M is set to [5, 10], and N=0.5M.

Images