Burnett, Kevin, authorDraper, Bruce, advisorRay, Indrakshi, advisorVijayasarathy, Leo, committee member2007-01-032007-01-032012http://hdl.handle.net/10217/67998Activity Recognition (AR) systems interpret events in video streams by identifying actions and objects and combining these descriptors into events. Relational databases can be used to model AR systems by describing the entities and relationships between entities. This thesis presents a relational data model for storing the actions and objects extracted from video streams. Since AR is a sequential labeling task, where a system labels images from video streams, errors will be produced because the interpretation process is not always temporally consistent with the world. This thesis proposes a PostgreSQL function that uses the Viterbi algorithm to temporally smooth labels over sequences of images and to identify track windows, or sequential images that share the same actions and objects. The experiment design tests the effects that the number of sequential images, label count, and data size has on execution time for identifying track windows. The results from these experiments show that label count is the dominant factor in the execution time.born digitalmasters thesesengCopyright and other restrictions may apply. User is responsible for compliance with all applicable laws. For information about copyright law, please see https://libguides.colostate.edu/copyright.PostgreSQLViterbiText