Data association using multisensor fusion

Data association is one of the most fundamental problems related to tracking; it is a means by which correct data is chosen to update the information on a target being tracked. Data association can also be viewed as a decision making process, where a decision is made as to which data point or which set of data points are to be used to update a tracking filter.; Previous data association algorithms, such as probabilistic data association filter and multiple hypothesis filter, to name a few, utilize only numerical data. As mentioned by Nahin and Pokoski (22), decisions based upon purely numerical data can be ambiguous. Due to the inherent ambiguity associated with numerical data, techniques for data association that utilize purely numerical data have limited the effectiveness. Blackman (21) and Nahin and Pokoski (22) allude to the use of non-numerical data and/or the use of Dempster-Shafer theory of mathematical evidence, yet they do not provide a means or a method of actually applying such data for data association. A new data association technique which utilizes fused data, consisting of both numerical and non-numerical attributes, is presented here. Fused data provide a compact and a useful representation of the target under consideration. Decisions based upon such data are less ambiguous, hence make the proposed technique very effective.; Numerical data are used to characterize a target's trajectory; whereas, non-numerical data identify the target's type. Such data may be obtained from multiple sensors and they are fused using the process of seeded clustering and the Dempster-Shafer theory of mathematical evidence. The use of fused data provide additional information regarding the target being tracked--hence, making the presented technique of data association more effective. The method presented not only provides the correct information to update a tracking filter but also facilitates the problem of track initiation and extends itself easily to multiple target tracking.; A tracking algorithm is developed that embeds the method presented for data association. This tracking algorithm which was designed specifically to handle numerical as well as non-numerical data, includes algorithms for data fusion, and is capable of track initiation, multiple target tracking, and track termination.