Template co-updating in multi-modal human activity recognition systems.