Researchers in pervasive computing have worked for decades on sensor-based human activity recognition (HAR). Among the digital health applications, the recognition of activities of daily living (ADL) in smart home environments enables the identification of behavioral changes that clinicians consider as a digital bio-marker of early stages of cognitive decline. The real deployment of sensor-based HAR systems in the homes of elderly subjects poses several challenges, with privacy and ethical concerns being major ones. This paper reports our experience applying privacy by design principles to develop and deploy one of these systems.