Finding Top-kappa Unexplained Activities in Video

Abstract : Most past work on identifying unexpected activities in video has focused on looking for specific patterns of anomalous activities. In this paper, we consider the situation where we have a known set A of activities (normal and abnormal) that we wish to monitor. However, in addition, we wish to identify abnormal activities that have not been previously considered or encountered, i.e. they are not in A. We formally define the probability that a video sequence is unexplained (totally or partially) w.r.t. A. We develop efficient algorithms to identify the top-k Totally and Partially Unexplained Activities in a video w.r.t. A. Our algorithms use neat mathematical properties of the definitions for efficiency. We describe experiments using two real-world datasets showing that our approach works well in practice in terms of both running time and accuracy.