Modelling Temporal Structures in Video Event Retrieval using an AND-OR Graph
conference paper
One of the challenges in Video Event Retrieval, the field in which (a sequence of frames with) high-level events are retrieved from a set of videos, is to model the temporal structure. One way to incorporate this information is using ANDOR graphs, which is a type of graphical model consisting of layers with AND nodes and OR nodes. We introduce new nodes, such as the BEFORE and WHILE node, for AND-OR graphs
to explicitly model temporal information. The advantage of these nodes is that the graph is insightful and transparent for a user. Additionally, the graph can both be created by a user or with the use of training examples. We perform initial experiments on a video surveillance dataset named VIRAT, which contains temporally inverse events with the same concepts, such as entering and exiting a building. We compare performance to state of the art Support Vector Machine and Hidden Markov Model methods. We show that our proposed graph with WHILE and BEFORE nodes outperforms the state of the art methods.
to explicitly model temporal information. The advantage of these nodes is that the graph is insightful and transparent for a user. Additionally, the graph can both be created by a user or with the use of training examples. We perform initial experiments on a video surveillance dataset named VIRAT, which contains temporally inverse events with the same concepts, such as entering and exiting a building. We compare performance to state of the art Support Vector Machine and Hidden Markov Model methods. We show that our proposed graph with WHILE and BEFORE nodes outperforms the state of the art methods.
TNO Identifier
868827
Source title
MMEDIA 2017 : The Ninth International Conferences on Advances in Multimedia
Files
To receive the publication files, please send an e-mail request to TNO Repository.