From Virtual to Real World: Applying Animation to Design the Activity Recognition System

摘要: Following the conventional pipeline, training dataset of a human activity recognition system relies on detection significant signal variation regions. Such position-specific classifiers provide less flexibility for users to alter sensor positions. In this paper, we proposed employ simulated generate corresponding from motion animation as dataset. Visualizing items real world, user can determine sensor’s placement arbitrarily and obtain accuracy feedback well classifier interface get relief cost model. With cases validation, trained by data effectively recognize real-world activity.