Study On Multisensor Collaborative Tracking In Sensor Networks

With the development of sensor and communication technologies, more and moresensors are incorporated in the integrated network to participate in the cooperativecombat. Multisensor information fusion which has the advantages of increasing thedimensions of measurement space and improving the accuracy of target tracking hasbecome a hot topic in academia and engineering. However, as the numbers of sensorand target increase, the complexity of information fusion system is increased. As aresult, the fusion structure needs to be optimized urgently. Moreover, the sensors whichcan be used for target tracking are limited, due to the variability and limitation of thetarget environment. Thus, how to distribute the limited sensor resources to collaborativetrack different targets has become a key problem of information fusion.In this paper, in order to improve the overall tracking accuracy of sensor networks,we focus on the collaborative tracking problem in resource constrained sensor networksbased on the previous work and the scientific research project. The main work can besummarized as follows:1. The basic theory and the mathematics derivation of collaborative trackingproblem are described. A general research framework of multisensor collaborativetracking is proposed based on the analysis of system dynamics model, observationmodel, glint noise of the measurement and the common filter algorithms. Thisframework shows that the main differences between collaborative tracking and generaltracking are the measurement of tracking accuracy as well as decision-making andexecution of sensor actions.2. The maneuvering target collaboration tracking in static sensor networks isresearched. Aiming to solve the problems of collaborative tracking within a networkcomposed of few and fixed sensors, and one sensor allowed working at each moment, anew maneuvering target collaborative tracking method based on Rényi information gainis proposed. Under the consideration of the tracking accuracy as the main factor, theparticle filtering algorithm is applied to obtain the Rényi information gain of eachsensor. And the sensor is selected according to the maximal Rényi information gain.Moreover, the kinematics state of the maneuvering target is estimated by interactingmultiple model method. The proposed method can select suitable sensor for targetadaptively and obtain a better performance compared with traditional methods.3. The maneuvering target collaboration tracking in dynamic sensor networks is researched. In large dynamic sensor networks, since sensors are numerous and theirpositions are unfixed, the energy consumption of information transfer in the networkcannot be ignored when multiple sensors working simultaneously. To tackle thisproblem, a maneuvering target collaborative tracking method with dynamic sensorsdeployment optimization is proposed. Firstly, the particle swarm optimization (PSO)algorithm is applied to optimize dynamic sensors location, and increase the effectivecoverage of the surveillance region while reduce sensor networks’ energy consumption.Then, the tracking sensors are selected according to the maximal Rényi informationgain and the minimal energy consumption by binary particle swarm optimizationmethod. Finally, the kinematic state of the maneuvering target is estimated byinteracting multiple model particle filtering algorithm, and the estimated states of theselected tracking sensors are fused. Simulation results show that the proposed methodcan improve the effective utilization rate of network resources.4. The multiple maneuvering targets collaboration tracking is researched. In thecase of multiple maneuvering targets, aiming at the issue of how to distribute the rightsensor to the right target with constrained sensor resource, a new collaborative trackingmethod based on game theory is proposed. Firstly, the interacting multiple modelextended particle filtering algorithm is applied to obtain the estimated state by eachsensor in the networks. Secondly, when the tracking accuracy is updated during targettracking, a new negotiation will start to reallocate resources. The negotiation willallocate more sensors to the negotiation caller for its tracking under the condition ofprotecting the interests of both sides. Finally, the estimated states of the selectedtracking sensors which belong to one target are fused. The proposed method canadaptively redistribute the right sensors to the right targets according to the system tasktimely and maintain a high tracking accuracy of all targets within sensor networks.5. The multi-target collaboration detection and tracking is researched. In actualmulti-target environment, since the number of targets in a surveillance region is timevarying, assignment scheme needs to be updated. To address the problem of sensorassignment according to the detection and tracking requirement of different targets inthe sensor resource constrained network, we propose a novel multi-target collaborativedetection and tracking algorithm based on posterior Cramér-Rao lower bound (PCRLB)and binary particle swarm optimization (BPSO). In the proposed method, the trackingaccuracy is measured by PCRLB which is independent of the filtering algorithmemployed, while a group of randomly generated particles are used to obtain thedetection probability of newborn targets. Moreover, BPSO is adopted to search the surveillance region for the optimal activated sensor subset which can maximize thetracking accuracy of existing tracks and the detection probability of newborn targets.Finally, the results of particle filtering of the selected tracking sensors are fused. Theproposed method can not only quickly identify newborn targets, but also optimize thetracking performance of the existing targets. Additionally, based on the fact that theRadar Cross-Section (RCS) of stealth target has a directional characteristic, we applyour method to collaborative tracking for stealth targets. Simulation results verify theeffectiveness of the method.