A multi-sensor target association method based on the framework of non-mutually exclusive evidence theory

Abstract

The present invention discloses a multi-sensor target association method based on the theoretical framework of non-mutually exclusive evidence. The method designs a basic probability allocation module based on the dual characteristics of the target. By constructing the basic probability number of the target association relationship evidence from the target spatial position distance and the difference in the target information source, the different attribute information of the multi-sensor target is comprehensively utilized. In addition, in order to quantify the non-complete mutual exclusion between different target objects to be associated, the present invention also designs a multi-source non-mutually exclusive evidence fusion rule based on the extended ECR-PCR rule, constructs the non-mutual exclusion through the similarity of the spatial position distance between the targets, and fuses the association relationship evidence of different targets based on the non-mutual exclusion. The present invention takes into account the non-uniqueness of the multi-sensor target association objects and has good association performance for multi-sensor targets.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-sensor target fusion and tracking, and in particular relates to a multi-sensor target association method. Background Art

[0002] Multi-sensor target association is a technique for determining the origin of a target across multiple sensors. It is a prerequisite and key to multi-sensor multi-target fusion and tracking. As the number of targets and the uncertainty of their motion characteristics in a scene continue to increase, the association relationships between multi-sensor targets become more ambiguous and complex, making target association difficult and hindering subsequent multi-sensor target fusion and tracking. Existing multi-sensor target association methods are divided into probabilistic modeling and data-driven approaches. Probabilistic modeling methods use prior information to construct an initial target motion model and then probabilistically match the target information to be associated with the initial target motion model to achieve multi-sensor target association. This type of method relies too much on prior information about the scene and targets, and therefore lacks effective association performance in scenarios with uncertain and variable targets. Data-driven approaches analyze and combine the various attributes of the targets to be associated, such as spatial position, velocity, orientation, and type, and utilize techniques such as clustering, probabilistic statistics, and neural networks to automatically match the multi-sensor targets to be associated. Compared to probabilistic modeling methods, data-driven approaches require no prior information and are more adaptable to different scenarios.

[0003] Multi-sensor target association within the framework of evidence theory specifically involves representing the association relationships between targets as elements in an identification framework. Based on information such as the spatial distance between targets and differences in recognition type, probabilities are assigned to the elements representing the corresponding target association relationships. The Basic Probability Assignments (BPAs) representing the association relationships between different targets are then fused as multi-source evidence. Through probabilistic transformation, the fused probabilities are converted into the confidence levels of each hypothesis element in the identification framework to achieve target association decisions. Although within the framework of evidence theory, the association relationships between multi-sensor targets are converted into the BPAs of different elements in the identification framework, improving the confidence level of multi-sensor target association decisions through information redundancy in the evidence of association relationships between different targets, target association decisions are based on the element with the highest confidence level in the identification framework, resulting in a unique target association object and potentially missed associations among multi-sensor targets. Therefore, based on the theoretical framework of non-mutually exclusive evidence, the present invention quantifies the non-uniqueness of target-related objects into non-mutually exclusive degrees for fusion in the multi-source evidence fusion stage representing different target association relationships, and directly performs target association decisions on the power set elements of the identification framework, which can greatly reduce the missed association rate of multi-sensor targets.

[0004] The scheme of multi-sensor target association technology under the existing evidence theory framework is as follows:

[0005] (1) Evidence modeling is performed on the association relationship between the target and other targets to be associated, and the potential target association relationship of each target is represented as the basic probability distribution under the corresponding identification framework.

[0006] (2) Using the spatial location distance between targets, information source differences and other attributes, a basic probability number is constructed to generate the evidence of the association relationship between the corresponding targets, indicating the confidence level of association, non-association and unknown association between the targets.

[0007] (3) The evidence after basic probability distribution corresponding to different target association relationships is fused to obtain the association probability between the target and all targets to be associated, thereby realizing the decision-making of multi-sensor target association.

[0008] The differences between different multi-sensor target association methods based on the evidence theory framework lie in the basic probability number generation model of the target association relationship evidence form in step (2) and the fusion rules of different target association relationship evidence in step (3). The basic probability number generation model is divided into the Antagonist model and the Non-Antagonist model according to whether the target association relationship has inherent conflict. The former believes that the confidence of association and non-association between targets can be non-zero at the same time, while the latter believes that the confidence of association and non-association between targets cannot be zero at the same time. The target association relationship evidence fusion rules include the Dempster rule and the PCR rule. The difference between them lies in the different ways of handling conflicting evidence. The Dempster rule fuses evidence based on the idea of ​​global normalization of non-conflicting evidence, while the PCR rule redistributes the basic probability numbers of conflicting evidence.

[0009] In the multi-sensor target association problem, the same target may be associated with multiple other targets simultaneously, meaning that the multi-sensor target association objects are non-unique. Existing methods assume that the target objects to be associated are completely mutually exclusive during the target association evidence fusion stage. This fails to account for the non-mutual exclusivity of target association evidence due to the non-uniqueness of target association objects during the evidence fusion stage, resulting in a certain degree of missed target associations and poor multi-sensor target association performance. Summary of the Invention

[0010] To overcome the shortcomings of the prior art, the present invention provides a multi-sensor target association method based on a non-mutually exclusive evidence theory framework. This method designs a basic probability allocation module based on the dual characteristics of the target. By constructing the basic probability number of target association relationship evidence from the target spatial position distance and the difference in target information source, it comprehensively utilizes the different attribute information of multi-sensor targets. In addition, in order to quantify the non-complete mutual exclusion between different target objects to be associated, the present invention also designs a multi-source non-mutually exclusive evidence fusion rule based on the extended ECR-PCR rule. The non-mutual exclusion is constructed through the similarity of the spatial position distance between the targets, and the association relationship evidence of different targets is fused based on the non-mutual exclusion. The present invention takes into account the non-uniqueness of the multi-sensor target association objects and has good association performance for multi-sensor targets.

[0011] The technical solution adopted by the present invention to solve the technical problem includes the following steps:

[0012] Step 1: Evidence modeling of the association relationship between the target under the multi-sensor and other targets to be associated is performed. The modeling method is as follows;

[0013] Step 1-1: For any two targets, considering the uncertainty of the relationship between them, an identification framework Ω = {y, n} is constructed based on evidence theory to represent the relationship between the targets, where y represents the relationship between the two targets and n represents the relationship between the two targets.

[0014] Step 1-2: Based on the recognition framework Ω constructed in step 1-1, the association relationship between the two targets is expressed as the following BPA form:

[0015] m:{m(y),m(n),m(y,n)}

[0016] In this BPA, m(y) represents the confidence that two targets are associated, m(n) represents the confidence that two targets are not associated, and m(y,n) represents the confidence that the association between two targets is unknown;

[0017] Step 2: In the recognition framework Ω, the confidence level of the association between two targets is calculated based on the two attributes of the target spatial location distance and the difference in information sources. The calculation method is as follows:

[0018] Step 2-1: The association relationship between two targets is related to their spatial position distance. Assuming that the distance between the targets is d, the confidence levels of the association relationship between them are:

[0019]

[0020] Among them, m p(·) represents the confidence of the spatial position distance between the association relationship between targets, α∈[0,1] represents the reliability of the sensor's detection of target information, and I represents the spatial position similarity between targets, which is calculated by the distance d and the threshold D:

[0021]

[0022] Step 2-2: The correlation between two targets also depends on whether they originate from the same sensor. Since a sensor detects each target only once at a given moment, two targets originating from the same sensor are necessarily uncorrelated. The confidence level of the correlation between targets with different information sources is calculated as follows:

[0023]

[0024] Among them, w s Indicates the confidence level when the target comes from different sensors and the correlation relationship is unknown, which is used to quantify the uncertainty of the difference in target information sources;

[0025] Step 2-3: Associate the spatial position distance between targets with the confidence m p (·) Confidence of the relationship between the difference and information source m s (·) is fused, and the fusion method adopts Dempster combination rule:

[0026]

[0027] Dempster fusion operator. The confidence of the relationship between the fused targets is as follows:

[0028]

[0029] Step 3: Considering the non-uniqueness of multi-sensor target association objects and the non-mutual exclusivity of identification framework elements in the non-mutually exclusive evidence theory, the association relationship BPA between different targets is extended to the non-mutually exclusive identification framework Θ i,. ={X (i,1) ,…,X (i,j) ,…,X (i,M) ,X (i,*)}Next;Θ i,. represents the set of association relationships between target i and other targets, M represents the total number of targets to be associated, X (i,j) Indicates the association relationship between target i and target j to be associated, X (i,*) Indicates that target i is not associated with any target; for the association relationship X between target i and target j to be associated (i,j) , whose confidence levels include:

[0030] (1) The confidence m of the association between target i and target j to be associated ij (y);

[0031] (2) Confidence m that target i is not associated with target j ij (n), which is equivalent to the confidence m associated with target i and all other targets i,(1,…,j-1,j+1,…M) (y);

[0032] (3) The confidence level m of the unknown relationship between target i and target j ij (y,n);

[0033] For the association relationship set Θ between target i and other targets i,. ={X (i,1) ,…,X (i,j) ,…,X (i,M) ,X (i,*)}, where each element X (i,j) The BPA includes the confidence m related to other goals i,(1,…,j-1,j+1,…M) (y); Therefore, by fusion of evidence in a non-mutually exclusive identification framework, the redundant information between elements can be fully utilized to improve the decision confidence of multi-sensor target association; the specific steps are as follows:

[0034] Step 3-1: Based on the spatial distance d between target i and target j (j=1,…,M) to be associated ij , construct non-mutually exclusive degree u i ; The calculation formula of non-mutual exclusion is as follows:

[0035]

[0036]

[0037] Among them, F a and F b Represents the potential relationship between target i and targets a and b respectively. When the spatial position similarity between targets a and b is I a,b The larger the value, the greater the possibility that two targets are associated with target i at the same time. a and F b The greater the non-mutual exclusion between them;

[0038] Step 3-2: According to the non-mutual exclusion u i , the basic probability number m of the conflict part of the evidence of the association relationship between target i and the target to be associated j, j = 1, ..., M is divided into two parts: the mutually exclusive part (1-u i )·m and non-mutually exclusive part u i ·m;

[0039] Step 3-3: The basic probability numbers of the mutually exclusive parts are proportionally distributed to the propositions corresponding to each sub-evidence; the basic probability numbers of the mutually exclusive parts are proportionally distributed to the propositions corresponding to each sub-evidence; The calculation process:

[0040]

[0041] Step 3-4: The basic probability numbers of the non-mutually exclusive parts are assigned to the union of the conflicting propositions; the basic probability numbers of the non-mutually exclusive parts are assigned to the union of the conflicting propositions; The calculation process:

[0042]

[0043] Step 3-5: Add the basic probability numbers of the two parts to obtain the complete non-mutually exclusive evidence fusion basic probability number

[0044]

[0045] Step 4: The fused association evidence M generated from the target i, i=1,…,M generated in the above steps i As shown below:

[0046]

[0047] Select the combination corresponding to the maximum confidence level to make an association decision:

[0048]

[0049] Preferably, the w s Set to 0.

[0050] The beneficial effects of the present invention are as follows:

[0051] The present invention basically does not rely on scene and target prior information. In scenarios where the number of targets and motion characteristics are uncertain, it takes into account the non-mutual exclusivity of different target association relationship evidence caused by the non-uniqueness of multi-sensor target association objects, achieves excellent multi-sensor target association performance, and reduces the target missed association rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is the overall framework diagram of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below with reference to the accompanying drawings and examples.

[0054] The purpose of the present invention is to propose a multi-sensor target association method based on the non-mutually exclusive evidence theory framework under the scenario constraints of uncertain target number and motion characteristics. Taking into account the non-uniqueness of multi-sensor target association objects, the non-complete mutual exclusion between different target objects to be associated is quantified, and the effective association of multi-sensor targets is achieved in the absence of prior information on the scene and targets.

[0055] like Figure 1 As shown in FIG, a multi-sensor target association method based on the non-mutually exclusive evidence theory framework includes the following steps:

[0056] Step 1: Evidence modeling is performed on the association relationship between the target under the multi-sensor and other targets to be associated. The modeling method is:

[0057] Step 101: For any two targets, considering the uncertainty of the association relationship between them, an identification framework Ω = {y, n} representing the association relationship between the targets is constructed based on evidence theory, where y represents the association between the two targets and n represents the non-association between the two targets.

[0058] Step 102: Based on the recognition framework Ω constructed in step 101, the association relationship between the two targets is expressed in the following BPA form:

[0059] m:{m(y),m(n),m(y,n)}

[0060] In this BPA, m(y) represents the confidence that two targets are associated, m(n) represents the confidence that two targets are not associated, and m(y,n) represents the confidence that the association between two targets is unknown;

[0061] Step 2: In the recognition framework Ω, the confidence level of the association between two targets is calculated based on the two attributes of the target spatial location distance and the difference in information sources. The calculation method is as follows:

[0062] Step 201: The association relationship between two targets is related to their spatial position distance. Assuming that the distance between the targets is d, the confidence levels of the association relationship between them are:

[0063]

[0064] Among them, m p (·) represents the confidence of the spatial position distance between the association relationship between targets, α∈[0,1] represents the reliability of the sensor in detecting target information, and I represents the spatial position similarity between targets, which can be calculated from the distance d and the threshold D:

[0065]

[0066] Step 202: The association between two targets also depends on whether they originate from the same sensor. Since a sensor detects each target only once at a given moment, two targets originating from the same sensor are necessarily unassociated. Based on this concept, the confidence level of the association between targets with different information sources is calculated as follows:

[0067]

[0068] Among them, w s Indicates the confidence level when the target comes from different sensors and the correlation relationship is unknown. It is used to quantify the uncertainty of the difference in target information sources and can generally be set to 0.

[0069] Step 203: Associating the spatial position distance between targets with the confidence m p (·) Confidence of the relationship between the difference and information source m s (·) is fused, and the fusion method adopts Dempster combination rule:

[0070]

[0071] Dempster fusion operator. The confidence of the relationship between the fused targets is as follows:

[0072]

[0073] Step 3: Considering the non-uniqueness of multi-sensor target association objects and the non-mutual exclusivity of identification framework elements in the non-mutually exclusive evidence theory, the association relationship BPA between different targets is extended to the non-mutually exclusive identification framework Θ i,. ={X (i,1), …,X (i,j) ,…,X (i,M) ,X (i,*)}Next. i,. represents the set of association relationships between target i and other targets, M represents the total number of targets to be associated, X (i,j) Indicates the association relationship between target i and target j to be associated, X (i,*) Indicates that target i is not associated with any target. For the association relationship X between target i and target j to be associated (i,j) , whose confidence levels include:

[0074] (1) The confidence m of the association between target i and target j to be associated ij (y)

[0075] (2) Confidence m that target i is not associated with target j ij (n), which is equivalent to the confidence m associated with target i and all other targetsi,(1,…,j-1,j+1,…M) (y)

[0076] (3) The confidence level m of the unknown relationship between target i and target j ij (y,n)

[0077] For the association relationship set Θ between target i and other targets i,. ={X (i,1) ,…,X (i,j) ,…,X (i,M) ,X (i,*)}, where each element X (i,j) The BPA includes the confidence m related to other goals i,(1,…,j-1,j+1,…M) (y). Therefore, by fusion of evidence under the non-mutually exclusive identification framework, we can make full use of the redundant information between elements and improve the decision confidence of multi-sensor target association. The specific steps are as follows:

[0078] Step 301: Based on the spatial distance d between target i and target j (j=1,…,M) to be associated ij , construct non-mutually exclusive degree u i The calculation formula of non-mutual exclusion is as follows:

[0079]

[0080]

[0081] Among them, F a and F b Represents the potential relationship between target i and targets a and b respectively. When the spatial position similarity between targets a and b is I a,b The larger the value, the greater the possibility that two targets are associated with target i at the same time. a and F b The greater the degree of non-mutual exclusion between them.

[0082] Step 302: According to the non-mutual exclusion degree u i , the basic probability number m of the conflict part of the evidence of the association relationship between target i and the target to be associated j (j=1,…,M) is divided into two parts: the mutually exclusive part (1-u i )·m and non-mutually exclusive part u i ·m.

[0083] Step 303: The basic probability numbers of the mutually exclusive parts are proportionally distributed to the propositions corresponding to each sub-evidence. The calculation process:

[0084]

[0085] Step 304: The basic probability numbers of the non-mutually exclusive parts are assigned to the union of the conflicting propositions. The calculation process:

[0086]

[0087] Step 305: Add the basic probability numbers of the two parts to obtain the complete non-mutually exclusive evidence fusion basic probability number

[0088]

[0089] Step 4: The association relationship evidence M after fusion of target i (i=1,…,M) generated in the above steps i As shown below:

[0090]

[0091] On this basis, the combination corresponding to the maximum confidence is selected for association decision:

[0092] Specific embodiment:

[0094] A simulation scenario with 5 targets and 4 sensors is set to simulate the target detection process of multiple sensors in an actual scenario. Simulation data such as the two-dimensional coordinate values ​​of the targets and the source sensors are generated. The method of the present invention is explained by taking the application of the method on the simulation data set as an example:

[0095] Step 1: 4 sensors generate approximately 11 detection targets for 5 actual targets. Each target has corresponding information such as coordinate value x, coordinate value y, detection sensor ID, target number under the detection sensor, and actual target ID. Evidence modeling is performed on the association relationship between these 11 multi-sensor detection targets to be associated. The modeling method is as follows:

[0096] Step 101: For any two targets i and j, considering the uncertainty of the relationship between them, an identification framework Ω is constructed based on evidence theory to represent the relationship between the targets. i,j ={y,n}, where y represents the association between two targets and n represents the non-association between two targets;

[0097] Step 102: Based on the recognition framework Ω constructed in step 101 i,j , the association relationship between the two targets is expressed as the following BPA form:

[0098] m ij :{m ij (y),m ij (n),mij (y,n)}

[0099] In this BPA, m ij (y) represents the confidence of the association between two targets, m ij (n) represents the confidence that the two targets are not associated, m ij (y,n) represents the confidence level of the unknown relationship between two targets;

[0100] Step 2: In the identification frame Ω ij In [1], the confidence of the association between two targets is calculated based on the two attributes of target spatial location distance and information source difference. The calculation method is as follows:

[0101] Step 201: The association relationship between two targets is related to their spatial position distance. Assume that the distance between the targets is d ij , then the confidence levels of their associations are:

[0102]

[0103] in, The confidence level of the spatial position distance that represents the relationship between targets. The reliability of the sensor in detecting target information is set to 0.9. ij Indicates the spatial position similarity between targets, which can be represented by the distance d ij The threshold D is calculated and the threshold D is set to 30, which means that the maximum detection error dimension of the sensor to the target position is 30m:

[0104]

[0105] Step 202: The association between two targets also depends on whether they originate from the same sensor. Since a sensor detects each target only once at a given moment, two targets originating from the same sensor are necessarily unassociated. Based on this concept, the confidence level of the association between targets with different information sources is calculated as follows:

[0106]

[0107] Among them, w s Indicates the confidence level when the target comes from different sensors and the correlation relationship is unknown. It is used to quantify the uncertainty of the difference in target information sources and can generally be set to 0.

[0108] Step 203: Associating the spatial position distance between targets with confidence Confidence of the relationship between differences in information sources Fusion is performed using the Dempster combination rule:

[0109]

[0110] Dempster fusion operator. The confidence of the relationship between the fused targets is as follows:

[0111]

[0112] Taking target 1 and target 2 as an example, the confidence of the relationship between their spatial position distance is Confidence of the relationship between information source differences To perform the fusion:

[0113]

[0114]

[0115] The Dempster rule is used to fuse the two BPAs mentioned above, and the fusion result is:

[0116]

[0117]

[0118]

[0119] The result indicates that the confidence level that target 1 and target 2 are associated is 0.32, the confidence level that they are not associated is 0.64, and the confidence level that the association relationship is unknown is 0.04.

[0120] Step 3: Considering the non-uniqueness of multi-sensor target association objects and the non-mutual exclusivity of identification framework elements in the non-mutually exclusive evidence theory, the association relationship BPA between the 11 targets is extended to the non-mutually exclusive identification framework Θ i,. ={X (i,1) ,…,X (i,j) ,…,X (i,11) ,X (i,*)},i=1,…,11。Θ i,. represents the set of association relationships between target i and other targets. The total number of targets to be associated is 11. (i,j) Indicates the association relationship between target i and target j to be associated, X (i,*) Indicates that target i is not associated with any target. For the association relationship X between target i and target j to be associated (i,j) , whose confidence levels include:

[0121] (1) The confidence m of the association between target i and target j to be associated ij (y)

[0122] (2) Confidence m that target i is not associated with target j ij (n), which is equivalent to the confidence m associated with target i and all other targets i,(1,…,j-1,j+1,…11) (y)

[0123] (3) The confidence level m of the unknown relationship between target i and target j ij (y,n)

[0124] For the association relationship set Θ between target i and other targets i,. ={X (i,1) ,…,X (i,j) ,…,X (i,11) ,X (i,*)}, where each element X (i,j) The BPA includes the confidence m related to other goals i,(1,…,j-1,j+1,…11) Therefore, by fusion of evidence under the non-mutually exclusive identification framework, the redundant information between elements can be fully utilized to improve the decision confidence of multi-sensor target association.

[0125] Taking three of the goals as an example, the specific steps are as follows:

[0126] Step 301: For target 1, according to the spatial position distance d between it and target 2 and target 3 12 and d 13 , and the spatial distance d between target 2 and target 3 23 , construct non-mutually exclusive degree:

[0127]

[0128]

[0129]

[0130] Step 302: According to the non-mutual exclusion degree u i , the basic probability number m of the conflict part of the evidence of the association relationship between target i and the target to be associated j (j=1,…,11) is divided into two parts: the mutually exclusive part (1-u i )·m and non-mutually exclusive part u i ·m.

[0131] Step 303: The basic probability numbers of the mutually exclusive parts are proportionally distributed to the propositions corresponding to each sub-evidence. The calculation process:

[0132]

[0133] Step 304: The basic probability numbers of the non-mutually exclusive parts are assigned to the union of the conflicting propositions. The calculation process:

[0134]

[0135] Step 305: Add the basic probability numbers of the two parts to obtain the complete non-mutually exclusive evidence fusion basic probability number

[0136]

[0137] Step 4: The fused association evidence M of target i (i=1,…,11) generated in the above steps i As shown below:

[0138]

[0139] On this basis, the combination corresponding to the maximum confidence is selected for association decision:

[0140]

Claims

1. A multi-sensor target association method based on the non-mutually exclusive evidence theory framework, characterized in that: The steps include: Step 1: Evidence modeling of the association relationship between the target under the multi-sensor and other targets to be associated is performed. The modeling method is as follows; Step 1-1: For any two targets, considering the uncertainty of the relationship between them, an identification framework Ω = {y, n} is constructed based on evidence theory to represent the relationship between the targets, where y represents the relationship between the two targets and n represents the relationship between the two targets. Step 1-2: Based on the recognition framework Ω constructed in step 1-1, the association relationship between the two targets is expressed as the following BPA form: m:{m(y),m(n),m(y,n)} In this BPA, m(y) represents the confidence that two targets are associated, m(n) represents the confidence that two targets are not associated, and m(y,n) represents the confidence that the association between two targets is unknown; Step 2: In the recognition framework Ω, the confidence level of the association between two targets is calculated based on the two attributes of the target spatial location distance and the difference in information sources. The calculation method is as follows: Step 2-1: The association relationship between two targets is related to their spatial position distance. Assuming that the distance between the targets is d, the confidence levels of the association relationship between them are: Among them, m p (·) represents the confidence of the spatial position distance between the association relationship between targets, α∈[0,1] represents the reliability of the sensor's detection of target information, and I represents the spatial position similarity between targets, which is calculated by the distance d and the threshold D: Step 2-2: The correlation between two targets also depends on whether they originate from the same sensor. Since a sensor detects each target only once at a given moment, two targets originating from the same sensor are necessarily uncorrelated. The confidence level of the correlation between targets with different information sources is calculated as follows: Among them, w s Indicates the confidence level when the target comes from different sensors and the correlation relationship is unknown, which is used to quantify the uncertainty of the difference in target information sources; Step 2-3: Associate the spatial position distance between targets with the confidence m p (·) Confidence of the relationship between the difference and information source m s (·) is fused, and the fusion method adopts Dempster combination rule: m=m p ⊕m s ⊕ represents the Dempster fusion operator. The confidence of the relationship between the fused targets is as follows: Step 3: Considering the non-uniqueness of multi-sensor target association objects and the non-mutual exclusivity of identification framework elements in the non-mutually exclusive evidence theory, the association relationship BPA between different targets is extended to the non-mutually exclusive identification framework Θ i,. ={X (i,1) ,…,X (i,j) ,…,X (i,M) ,X (i,*) }Next;Θ i,. represents the set of association relationships between target i and other targets, M represents the total number of targets to be associated, X (i,j) Indicates the association relationship between target i and target j to be associated, X (i,*) Indicates that target i is not associated with any target; for the association relationship X between target i and target j to be associated (i,j) , whose confidence levels include: (1) The confidence m of the association between target i and target j to be associated ij (y); (2) Confidence m that target i is not associated with target j ij (n), which is equivalent to the confidence m associated with target i and all other targets i,(1,…,j-1,j+1,…M) (y); (3) The confidence level m of the unknown relationship between target i and target j ij (y,n); For the association relationship set Θ between target i and other targets i,. ={X (i,1) ,…,X (i,j) ,…,X (i,M) ,X (i,*) }, where each element X (i,j) The BPA includes the confidence m related to other goals i,(1,…,j-1,j+1,…M) (y); Therefore, by fusion of evidence in a non-mutually exclusive identification framework, the redundant information between elements can be fully utilized to improve the decision confidence of multi-sensor target association; the specific steps are as follows: Step 3-1: Based on the spatial distance d between target i and target j (j=1,…,M) to be associated ij , construct non-mutually exclusive degree u i ; The calculation formula of non-mutual exclusion is as follows: Among them, F a and F b Represents the potential association relationship between target i and targets a and b respectively. When the spatial position similarity between targets a and b is I a,b The larger the value, the greater the possibility that two targets are associated with target i at the same time. a and F b The greater the non-mutual exclusion between them; Step 3-2: According to the non-mutual exclusion u i , the basic probability number m of the conflict part of the evidence of the association relationship between target i and the target to be associated j, j = 1, ..., M is divided into two parts: the mutually exclusive part (1-u i )·m and non-mutually exclusive part u i ·m; Step 3-3: The basic probability numbers of the mutually exclusive parts are proportionally distributed to the propositions corresponding to each sub-evidence; the basic probability numbers of the mutually exclusive parts are proportionally distributed to the propositions corresponding to each sub-evidence; The calculation process: Step 3-4: The basic probability numbers of the non-mutually exclusive parts are assigned to the union of the conflicting propositions; the basic probability numbers of the non-mutually exclusive parts are assigned to the union of the conflicting propositions; The calculation process: Step 3-5: Add the basic probability numbers of the two parts to obtain the complete non-mutually exclusive evidence fusion basic probability number Step 4: The fused association evidence M generated from the target i, i=1,…,M generated in the above steps i As shown below: Select the combination corresponding to the maximum confidence level to make an association decision:

2. The multi-sensor target association method based on the non-mutually exclusive evidence theoretical framework according to claim 1 is characterized in that: The w s Set to 0.

Images