This paper proposes an approach to analyze the temporal and spatial correlations between objective activities from multiple non-overlapping camera network.Based on the similarity of moving models and the relationship of moving space

each vision field of camera network is segmented into semantic active regions automatically.Then a Cross Kernel Canonical Correlation Analysis is implemented to explore the correlations between these active regions and the topology of the camera network.This topology can reflect the temporal and spatial information of objectives cross multi-cameras and improve the accuracy of objects re-identification by removing false objects.Compared with existing methods

our approach does not depend on the individual tracking and is efficient in complex and crowed scene.The experiment results show that our approach performs effectively and efficiently in multi camera surveillance network.