Latent Dirichlet Allocation with Soft Assignment of Descriptors to Words
Gal Levl · 2013
Automatic processing of video data is essential in order to allow efficient access to large amounts of video content, a crucial point in applications such as video mining and surveillance. In this paper we focus on the problem of identifying interesting parts of the video. Specifically, we seek to identify atypical video events, which are the events a human user is usually looking for. As is common in the literature, we equate atypical events to events of low-probability with respect to a model that describes normal events. We propose to identify atypical events by modeling a corpus of typical video events using the Latent Dirichlet Allocation model. Subsequently, to classify an event as atypical we compute its probability with respect to the learned LDA model. Furthermore, we’ve extended the LDA model, which works with discrete data, to work with continuous data as in our case of video events. We tested our algorithm on the UCSD ped2 [20] and UMN datasets [1], which were previously used to evaluate anomaly detection algorithms.