A Framework of Context-A ware Object Recognition for Smart Home

摘要

Services for smart home share a fundamental problem-object recognition, which is challenging because of complex background and appearance variation of object. In this paper we develop a framework of object recognition for smart home integrating SIFT (scale invariant feature transform) and context knowledge of home environment. The context knowledge includes the structure and settings of a smart home, knowledge of cameras, illumination, and location. We counteract sudden significant illumination change by trained support vector machine (SVM) and use the knowledge of home settings to define the region for multiple view registration of an object. Experiments show that the trained SVM can recognize and distinguish different illumination classes which significantly facilitate object recognition.