RECOGNIZING FOOD PREPARATION ACTIVITIES USING BAG OF FEATURES

Abstract

Food preparation activities for cooking in the kitchen involve physical interactions between multiple objects such as hands, utensils, and
ingredients. Recognizing these complex activities using sensors embedded in kitchen utensils is challenging. For accurate recognition, it is
necessary to design efficient feature representation of sensor data. In this paper, we propose a feature learning method based on bag of features for food preparation activity representation and recognition. The activity model is built using histograms of motion primitives. We experimentally
validate the effectiveness of the proposed approach for recognizing ten activity classes. The experiment results show that the proposed approach
provided substantially higher accuracy than traditional approaches for food preparation activity recognition using embedded sensors.