Study of Computational Structures for Multiobject Tracking Algorithms

摘要

The structure and performance of multiobject tracking algorithms are heavily constrained by their current implementation on sequential computer architectures. Near-optimal algorithms, such as the track-oriented approach under investigation here, impose prohibitive requirements on sequential processors. These requirements force one to use near-optimal algorithms only in low target density environments and to employ faster but less optimal tracking algorithms in the more demanding high density situations. Identifying, enhancing, and exploiting the parallel computational structures in the near-optimal tracking algorithms potentially allows them to be applied to a wider range of scenarios. This report describes research undertaken to determine how parallel computer architectures might be utilized to implement sophisticated modern tracking algorithms which are beyond the capability of current sequential processing hardware. (Author)