Identification of Cooking Preparation Using Motion Capture Data: A Submission to the Cooking Activity Recognition Challenge

The Cooking Activity Recognition Challenge tasked the competitors with recognizing food preparation using motion capture and acceleration sensors. This paper summarizes our submission to this competition, describing how we reordered the training data, relabeled it and how we handcrafted features for this dataset. Our classification pipeline first detected basic user actions (micro-activities); using them it recognized the recipe, and then used the recipe to refine the original micro-activity predictions. After the post-processing step using a Hidden Markov Model, we achieved the competition score of 95% on the training data with cross-validation.