Towards recognizing food preparation activities in situational support systems

One way in which technology can help people with cognitive impairments to stay independent for longer is through situational support systems that recognize a person’s activities and provide help if needed. While food preparation is one of the most important tasks of daily living, such activities are extremely challenging to recognize automatically as they involve a large number of different objects, gestures and cognitive capabilities. This paper introduces an approach to recognizing food preparation activities using statistical machine learning based on visual and accelerometer data. As a first step towards multi-modal activity recognition, an accelerometer localization algorithm has been developed that provides important information for fusing visual and accelerometer data. We discuss a statistical activity model that will be used to guide the construction of datasets for complex activities and give an outlook on the next steps.