Probabilistic spatio-temporal inference for motion event understanding

In Video data, moving object has temporal flow and spatial features, which can be expressed in spatio-temporal relation. The goal in this paper is the understanding of object movement of spatio-temporal relation through mapping between vocabulary and object movement. In this case, spatio-temporal relation consists of temporal relation obedient to the passage of time, directional relation obedient to changes of object movement direction, changes of object size relation, topological relation obedient to changes of object movement position, and velocity relation using concept relations between topology models. This paper in the ontology building defines the inference rules using the proposed spatio-temporal relation and the use of Markov Logic Networks for probabilistic reasoning. In the experiments, motion verbs are used to understand semantic object movement. Probability weight and learning for 10,000 times are used for value comparison. The result value from inference exists even though connection rules such as “go through” are not defined directly. In addition, it is indicated that the relation that includes large number of connections such as “go to” has the high value of probabilistic inference result and that small number of connection relations depending on the change in object size such as “include” leads to low value of result.