Efficient Activity Retrieval through Semantic Graph Queries Efficient Activity Retrieval through Semantic Graph Queries
Paper summary This paper poses the the problem of querying a large corpus of aerial video as a subgraph matching problem. Here the video data has been transformed into a large graph where each frame contains labeled objects such as person, object, or car which become nodes and then the edges are relationships such as time (between sequential video frames) and distance (in current frame and in future frames). The reason the graph is built is to we can query it with graphs that represent what we are looking for. The first example in the paper (below) shows an example query (called $Q$). This query asks to "Find a person near an object then after some time or distance they are still near and there is a car". ![](http://i.imgur.com/6AKVCYX.png) The goal now is to find the most similar subgraphs in the larger graph. The game here is to reduce the complexity of the search into something that is not as bad as the subgraph isomorphism problem. Even though this is worse because we what things that are similar and not necessarily exact to the query. They filter the larger graph (that represents the video) into a smaller graph that only includes nodes and edges that can match those in the query graph (This graph is called the coarse graph $C$). Another is to filter the query $Q$ into a smaller graph $T$ which retains nodes and edges that have the most discriminative power. ### WORK IN PROGRESS

Summary by Joseph Paul Cohen 4 years ago
Your comment:

ShortScience.org allows researchers to publish paper summaries that are voted on and ranked!

Sponsored by: and