PhD forum: Dempster-Shafer based camera contribution evaluation for task assignment in vision networks

Abstract

In a network of cameras, it is important that the right subset of cameras takes care of the right task. In this work, we describe a general framework to evaluate the contribution of subsets of cameras to a task. Each task is the observation of an event of interest and consists of assessing the validity of a set of hypotheses. All cameras gather evidence for those hypotheses. The evidence from different cameras is fused by using the Dempster-Shafer theory of evidence. After combining the evidence for a set of cameras, the remaining uncertainty about a set of hypotheses, allows us to identify how well a certain camera subset is suited for a certain task. Taking into account these subset contribution values, we can determine in an efficient way the set of subset-task assignments that yields the best overall task performance.