Detection and Classification of Sleep Posture at Edge Device in Smart Home

Sleep quality is a key attribute, especially for elders’ health checks and assessments. Assessing the sleep position is an important factor in determining sleep quality, and finding ways to identify the sleeping positions poses significant challenges in the field of elderly care. Sleep posture detection and classification have recently become a great research paradigm in a diverse range of smart home applications. Constant and autonomous sleep posture monitoring reduces health risks and provides substantial information on the patient’s sleeping pattern. Most of the existing works use physical sensing mechanisms attached to human body parts that inconvenience the patient with prolonged use. This work presents an automated computer vision-based approach to continuously monitoring and classifying sleep posture on edge devices that sends notifications on the detection of sensitive classes. Edge computing delivers better data security and privacy to users’ sensitive personal information. An optimized detection algorithm has been proposed by leveraging approximate computing to reduce the computation complexity. Further, the optimized model is deployed on Raspberry Pi with a night-vision camera and in-build WiFi module to validate the efficacy on resource-constrained edge computing devices. The proposed system achieves a better performance with a reduced model size of 5.2MB and a runtime of 6.3 milliseconds in the hardware environment.