Target tracking in a multisensor environment using neural networks
AuthorWong, Yee Chin
AdvisorSundareshan, Malur K.
MetadataShow full item record
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
AbstractThe advent of technology has brought to the field of engineering many tools that were once considered impractical. For example, the increased processing speed of microprocessors now allows measurements from image sensors to be used for target tracking or target identification in real time--a task once thought unachievable. Of late the advances made in artificial intelligence (AI), specifically the artificial neural network, have sprung many different applications among which the implementation of AI controllers being the most popular. However, these advances have been slow in their implementation in the field of target tracking for several reasons. First, there seems to be a lack of sound tracking architectures that can exploit the use of artificial intelligent agents. Second, there is some difficulty in fusing the different forms of information that can be measured from the various available sensors such as the image sensor, millimeter wave radar, Doppler radar, etc. Third, the increased computational complexity due to the employment of the various sensors could limit the practical usefulness of such a tracking system. This dissertation presents a novel framework in which various dissimilar sensors can be used simultaneously to track a highly agile and non-cooperative target. The proposed framework not only allows the usage of multiple sensors to yield a robust and accurate tracker but also maintain a reasonable computational requirement. Unlike the methods proposed in the literature for the design of multi-sensor tracking systems, this dissertation presents an AI-based system that can accept, process, and fuse measurements from any number of sensors of dissimilar forms. The principal contributions of this dissertation are the following: (i) a novel architecture of a three-layer feedforward neural-network-based tracking system with the ability to fuse measurements from dissimilar sensors; (ii) a powerful optimization algorithm for training the neural network; (iii) a novel mathematical target motion model to simplify the training and implementation of the proposed tracking system.
Degree ProgramGraduate College
Electrical and Computer Engineering