Abstract
Activity recognition, having endemic impact on smart homes, faces one of the biggest challenges in learning a personalized activity model completely by using a generic model especially for parallel and interleaved activities. Furthermore, inhabitant's mistaken object interaction may entail in another spurious activity at smart homes. Identifying and removing such spurious activities is another challenging task. Knowledge driven techniques used for recognizing activity models are static in nature, lack contextual representation and may not comprehend spurious actions for parallel/interleaved activities. In this paper, a novel approach for completing the personalized model specific to each inhabitant at smart homes using generic model (incomplete) is presented that can recognize the sequential, parallel, and interleaved activities dynamically while removing the spurious activities semantically. A comprehensive set of experiments and results based upon number of correct (true positivity) or incorrect (false negativity) recognition of activities assert effectiveness of presented approach within a smart home.