Fusion of Event Detection Systems: An Application to Keyword Search

Fusion of data obtained by different sensors or the systems that process them, in order to increase the accuracy of detection, is a widely applied method. In the event detection problems, on the other hand, the time-alignment and fusion of the hypotheses proposed by different systems becomes a computationally expensive task as the number of hypotheses increase. In this study, a new fusion methodology is proposed, which facilitates a fast alignment and fusion of hypotheses provided by different keyword search systems and, it was shown that the fusion system performs better than the best individual system.