Three methods for energy-efficient context recognition