A multi-detector collaborative tracking method based on a confidence model

Abstract

The present invention discloses a multi-detector collaborative tracking method based on a confidence model. In a complete system, it often does not exist in the form of a single detector. However, in many cases during use, all detectors are not organically combined, and a single detector working alone can gradually no longer fully meet the task requirements. Therefore, multi-detector collaboration provides an idea for completing more complex tracking tasks. Multi-detector collaborative work has two advantages. On the one hand, due to the advantages of the distributed working mode, the distributed collaborative tracking system has stronger robustness and will not cause the system to crash due to the damage of a single detector. On the second hand, each detector has different working characteristics, and the information obtained for the same target will be richer. The present invention designs a detector confidence model according to the detector type information and working conditions, effectively broadening the working field of the system.

Description

Technical Field

[0001] The present invention relates to the technical fields of detection and tracking, and distributed multi-agent collaboration, and particularly relates to a multi-detector collaborative tracking method based on a confidence model. Background Art

[0002] With the development of tracking technology, there are currently many algorithms for target tracking, and the complexity of tracking tasks is also increasing. The traditional method of completing tracking tasks through a single device can no longer meet the task requirements. The working mode of multi-device collaboration has gradually entered various fields and is also one of the important research directions at present.

[0003] Collaboration technology originates from distributed networks. In traditional networks, there is a central node, and all nodes send data to the central node for it to complete data calculation and processing. Such a network often has high working efficiency, and the network structure is relatively simple. Since the emergence of such a network, it has been widely used in various fields. However, such a centralized network structure has the problem of low robustness. Once the central unit of the network is damaged, the entire network may collapse. To solve this problem, distributed networks emerged. In a distributed network, the concept of a central unit is removed, and each network node can independently perform calculation and processing tasks on data. Such a network has stronger robustness than a centralized network and also has stronger scalability.

[0004] The core idea of a distributed network is collaboration and sharing. Each unit in the network shares data with its neighbor units, so that all nodes will have a large amount of network information, which helps the system complete tasks efficiently. In a target tracking system, the idea of collaboration can also be introduced. In current target tracking tasks, detectors are usually used to capture targets and obtain target information. The target data is then solved and processed in the system to obtain the motion information of the target, and the target is tracked through a transmission system. Therefore, in the entire system, the stable capture ability of the detector for the target is the first and important step to complete tracking. There are many types of detectors, and the working characteristics of different detectors will vary greatly. For example, for infrared radiation detectors, detectors in different bands have very different detection capabilities for different targets. In a large-scale detection and tracking system, detectors usually do not exist alone and are mostly in a state of multi-detector integration. Therefore, it is necessary to design relevant collaborative algorithms according to the working characteristics of detectors to improve the working ability of the entire system. In the present invention, when designing the collaboration mode, the working characteristics of detectors are fully considered, and the advantages of detectors are organically combined to improve the working ability of the system. Summary of the Invention

[0005] The object of the present invention is to provide a multi-detector collaborative tracking method based on a confidence model, which can be used in a detection and tracking system to effectively collaborate multiple detectors in the system, improving the system stability and tracking reliability.

[0006] The technical solution adopted by the present invention is as follows: A multi-detector collaborative tracking method based on a confidence model, the collaborative tracking algorithm fully considers information in two aspects: the detector type and the historical performance of the detector, designs a detector confidence scoring mechanism, and applies it to the distributed Kalman. This method has the following steps:

[0007] Step (1), Initialize the parameters of the collaborative tracking algorithm. When the collaborative tracking algorithm runs for the first time, use the initial values for subsequent calculations. In subsequent iterations, use the results of the previous iteration as the initial values for starting the calculations. In this step, the initialization of parameters ψ k,i and P k,i is involved;

[0008]

[0009] P k,i = P k,i|i-1

[0010] wherein, represents the state estimation quantity of the system at the previous moment, P k,i represents the state estimation covariance at the current moment, ψ k,i represents the intermediate estimation of the system, and this parameter will be updated in the subsequent steps;

[0011] Step (2), Establish a confidence model for the detector according to the detector type and the historical performance of the detector, score the detector, and then calculate the detector information fusion weight w k,i ;

[0012] Step (3), Incremental update process. The incremental process is the process of communication between detectors. Multiple detectors share data with each other, and at the same time each detector updates its local parameters;

[0013] Step (4), Data fusion. Through the above steps, each detector will obtain corresponding local data. At this time, all detectors share the local data in the network, and each detector fuses the received local data to obtain the final estimation quantity;

[0014] Step (5), The collaborative tracking algorithm updates the parameters according to the algorithm model P k,i+1|i ;

[0015] Step (6), The collaborative tracking algorithm enters the next iteration.

[0016] Further, in step (1), the cooperative tracking algorithm is an iterative algorithm. The characteristic of the iterative algorithm is that the result calculated in the previous round of iteration can be used as the starting point for the next round of iteration, and the optimal solution can be finally obtained through multiple rounds of iteration.

[0017] Further, in step (2), the establishment of the confidence model mainly designs the scoring mechanism for detection, and designs the confidence of the detector from two aspects as follows:

[0018] Confidence = Detector type + Detector historical performance

[0019] During the operation of the system, a scoring mechanism is designed. For different settings of the detector, a scoring mechanism for the detector is advanced. At the same time, according to the historical performance of the detector, a scoring mechanism is formulated as follows:

[0020] DeScore = SenType + HisScore

[0021] Among them, DeScore represents the confidence score finally obtained by the detection in the system, HisScore represents the score obtained from the historical performance of the detector, SenType is the detector type, and the detectors are divided into different types according to the detection ability: medium-wave detectors, long-wave detectors, and short-wave detectors.

[0022] The scoring mechanism is designed as follows:

[0023]

[0024] Among them,

[0025] e = ψ k,i -ψ k,i-1

[0026] w l,k = f(DeScore)

[0027] f(·) represents the system score conversion function.

[0028]

[0029] Among them, th represents the set average score, and SysTol represents the total system score.

[0030] Further, in step (3), the working mode of the distributed algorithm is relative to the centralized algorithm. In the working mode of the distributed algorithm, the concept of the central unit is removed, and all nodes obtain information by communicating with adjacent units. In this step, each detector will exchange data with adjacent detectors and update the current local estimate.

[0031] Further, in step (4), data fusion is the core of the distributed algorithm. Each detector will obtain the final estimate through data fusion. In this step, each detector needs to fuse the local estimation data of other adjacent detectors.

[0032] Further, in step (5), the Kalman algorithm has a relatively wide application in the field of tracking. The cooperative tracking algorithm uses Kalman as the basic algorithm model, corrects the estimated value through the observed quantity, and finally returns to the current state quantity. In this step, the update of parameters is mainly completed, mainly involving the parameters P k,i+1|i update.

[0033] Further, in step (6), the finally obtained state quantity is used as the starting point for the next round of iteration.

[0034] The advantages of the present invention compared with the prior art are as follows:

[0035] (1) The present invention introduces the cooperative idea into the detection and tracking system, improving the stability and working ability of the system.

[0036] (2) The present invention fully considers the working characteristics of the detectors, establishes a system confidence model, scores the detectors during the execution of the algorithm, and then adjusts the information fusion weights.

[0037] (3) The present invention establishes a system cooperative tracking working mechanism based on confidence, improving the working ability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flowchart of a multi-detector cooperative tracking algorithm based on a confidence model of the present invention;

[0039] Figure 2 shows a schematic diagram of the detection band information of common infrared radiation detectors;

[0040] Figure 3 is a data simulation diagram of the cooperative tracking algorithm based on the detector confidence. DETAILED DESCRIPTION OF THE INVENTION

[0041] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0042] Figure 1 is a flowchart of a multi-detector cooperative tracking algorithm based on a confidence model of the present invention, which shows the process of scoring the detectors according to the detector types and historical performances and performing weight conversion; Figure 2The detection band information of common infrared radiation detectors is given. Different types of detectors are often placed in the same device. The algorithm designed in the present invention can effectively coordinate multiple detectors organically and improve the working ability of the device;

[0043] The specific process steps of a multi-detector collaborative tracking method based on a confidence model according to the present invention are as follows:

[0044] Step (1), Initialize the algorithm parameters. In this step, ψ k,i and P k,i parameters are initialized;

[0045]

[0046] P k,i = P k,i|i-1

[0047] Among them, represents the state estimation quantity of the system at the previous moment, and P k,i represents the state estimation covariance at the current moment, and ψ k,i represents the intermediate estimation of the system, and this parameter will be updated in the subsequent steps;

[0048] Step (2), Establish a confidence model for the detector according to the type of the detector and the historical performance of the detector, score the detector, and then calculate the information fusion weight w k,i ;

[0049] The establishment of the confidence model mainly designs the scoring mechanism of detection. The confidence of the detector is designed from two aspects as follows:

[0050] Confidence = Detector type + Detector historical performance

[0051] During the operation of the system, a scoring mechanism is designed. For different detectors, we set the detector type score in advance, and at the same time, according to the historical performance of the detector, a scoring mechanism is formulated as follows:

[0052] DeScore = SenType + HisScore

[0053] Among them, DeScore represents the confidence score finally obtained by the detection in the system, HisScore represents the score obtained from the historical performance of the detector, and SenType is the detector type. The detectors are divided into different types according to the detection ability (such as medium-wave detectors, long-wave detectors, short-wave detectors):

[0054]

[0055] Among them,

[0056] e = ψ k,i -ψ k,i-1

[0057] w l,k = f(DeScore)

[0058] f(·) represents the system score conversion function,

[0059]

[0060] where th represents the set average score and SysTol represents the total system score.

[0061] Step (3), incremental update process. Multiple detectors share data through communication, and each detector updates its local parameters based on the acquired data;

[0062]

[0063]

[0064]

[0065] where R l,i represents the measurement noise covariance of detector l at time i, H l,i represents the measurement matrix of detector l at time i that converts the state quantity into the observable quantity, δ k,i represents the combination factor of detector k at time i, which can be obtained through step (2), y l,i represents the observed value of detector l at time i;

[0066] Step (4), data fusion. Each detector shares its local parameters in the distributed network and simultaneously acquires the local parameters of other sensors and performs data fusion;

[0067]

[0068] where N k represents the detectors that have data communication with detector k, which we call neighbor detectors, c l,k represents the weight of the algorithm, and the sum of the weights is 1:

[0069] Step (5), algorithm parameter P k,i+1|i update,

[0070] P k,i|i = P k,i

[0071]

[0072]

[0073] Among them, F i represents the state transition matrix at the current moment, G i represents the state noise transition matrix at the current moment, Q i represents the state noise covariance;

[0074] Step (6): Enter the next round of iteration.

[0075] Figure 3 is the data simulation diagram of the collaborative tracking algorithm based on the detector confidence. The upper curve (CoTrack) gives the algorithm accuracy of the collaborative work of the detectors, and the lower curve (TrustCoTrack) gives the algorithm accuracy under the used confidence model. It is found that the collaborative tracking algorithm designed according to the detector characteristics and historical performance can improve the algorithm accuracy.

[0076] The method of the present invention introduces the idea of distributed collaboration into the detection and tracking system, designs a working mode of multi-detector collaboration, and can effectively improve the stability and working ability of the system.

[0077] The above is only the specific implementation manner in the present invention, but the protection scope of the present invention is not limited thereto. Any transformation or replacement that can be understood and conceived by those familiar with the technology within the technical scope disclosed by the present invention should be covered within the scope of the present invention. The content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

Claims

1. A multi-detector collaborative tracking method based on a confidence model, characterized in that: The collaborative tracking algorithm fully considers the information of two aspects: the detector type and the historical performance of the detector, designs a detector confidence scoring mechanism, and applies it to the distributed Kalman. This method has the following steps: Step (1), Initialize the parameters of the collaborative tracking algorithm. When the collaborative tracking algorithm runs for the first time, use the initial values for subsequent calculations. In subsequent iterations, use the results of the previous iteration as the initial values for starting the calculations. In this step, ψ k,i and P k,i parameters are initialized; P k,i = P k,i|i-1 Among them, represents the state estimator of the system at the previous moment, P k,i represents the state estimation covariance at the current moment, ψ k,i represents the intermediate estimation of the system, and this parameter will be updated in the subsequent steps; Step (2): Establish a confidence model for the detector based on the type of the detector and its historical performance, score the detector, and then calculate the detector information fusion weight w k,i ; Step (3), Incremental update process. The incremental process is the process of communication between detectors. Multiple detectors share data with each other, and at the same time each detector updates its local parameters; Step (4), Data fusion. Through the above steps, each detector will obtain corresponding local data. At this time, all detectors share local data in the network, and each detector fuses the local data received to obtain the final estimate; Step (5), the collaborative tracking algorithm updates the parameters according to the algorithm model P k,i+1|i ; Step (6), The collaborative tracking algorithm enters the next round of iteration.

2. The multi-detector collaborative tracking method based on a confidence model according to claim 1, characterized in that: In step (1), the collaborative tracking algorithm is an iterative algorithm. The characteristic of the iterative algorithm is that the result calculated in the previous round of iteration can be used as the starting point for the calculation in the next round of iteration. Through multiple rounds of iteration, the optimal solution can be finally obtained.

3. The multi-detector collaborative tracking method based on a confidence model according to claim 1, characterized in that: In step (2), the establishment of the confidence model mainly designs the scoring mechanism for detection. The confidence of the detector is designed from two aspects as follows: Confidence = Detector type + Detector historical performance During the operation of the system, a scoring mechanism is designed. For different detectors, the detector scoring mechanism is set in advance. At the same time, according to the historical performance of the detector, a scoring mechanism is formulated as follows: DeScore = SenType + HisScore Among them, DeScore represents the confidence score finally obtained by the detection in the system, HisScore represents the score obtained from the historical performance of the detector, SenType is the detector type, and the detectors are divided into different types according to the detection ability: medium-wave detectors, long-wave detectors, short-wave detectors; The scoring mechanism is designed as follows: Among them, e = ψ k,i -ψ k,i-1 w l,k = f(DeScore) f(·) represents the system score conversion function, Among them, th represents the set average score, and SysTol represents the total system score.

4. The multi-detector collaborative tracking method based on a confidence model according to claim 1, characterized in that: In step (3), the working mode of the distributed algorithm is relative to the centralized algorithm. In the working mode of the distributed algorithm, the concept of the central unit is removed. All nodes obtain information by communicating with adjacent units. In this step, each detector will exchange data with adjacent detectors and update the current local estimate.

5. The multi-detector collaborative tracking method based on a confidence model according to claim 1, characterized in that: In step (4), data fusion is the core of the distributed algorithm. Each detector will obtain the final estimate through data fusion. In this step, each detector needs to fuse the local estimate data of other adjacent detectors.

6. The multi-detector collaborative tracking method based on a confidence model according to claim 1, characterized in that: In step (5), the Kalman algorithm has a relatively wide application in the field of tracking. The collaborative tracking algorithm uses Kalman as the basic algorithm model, corrects the estimated value through the observed quantity, and finally returns the current state quantity. In this step, the update of parameters is mainly completed, mainly involving the update of the parameter P k,i+1|i .

7. The multi-detector collaborative tracking method based on a confidence model according to claim 1, characterized in that: In step (6), the finally obtained state quantity is used as the starting point for the next round of iteration.

Images