In general, the development of prediction methods is a quite challenging field. However, as difficult the development is, as useful those methods can be in a large variety of use cases. Whether the weather of tomorrow or the destination of a moving individual is to be predicted, in both cases many different aspects have to be considered, because as well as the weather's behaviour the one of individuals, especially human beings, is influenced by many factors. For instance, movements of human beings are either planned, arbitrary or influenced by their environment or social aspects. In most cases a combination of those factors is involved. In this paper, motivated by the context of a decentralized surveillance scenario, we present an approach for predicting movements on the basis of a prediction model generated from the knowledge, which is implicated in spatio-temporal trajectories. This model is based on extracted interesting places and considers several aspects, which contain gained information about the movement behaviour in a given scenario.