Development of a Low-Power Microcontroller-Based Wrist-Worn Device With Resource-Constrained Activity Detection Algorithm

This article deals with the development of a resource-constrained activity detection algorithm for microcontroller-based wrist-worn devices. The various market available activity detection devices lack adequate agility and ease of use in certain applications, as they heavily rely upon computationally intensive algorithms to detect physical activity and step count. Therefore, by extension, they always need to stay connected with a smartphone that can perform these computations. To address this problem, this article introduces a standalone wearable device with the simplified algorithm for activity detection in a human being. The developed system consists of microelectromechanical systems (MEMS) triaxial accelerometer connected to a low-power microcontroller with a wrist anchor-able form factor. The step detection algorithm uses a moving average threshold filter to detect the change in acceleration data to identify specific events. A detailed comparative study considering power consumption and the step detection made between the developed device (with the resource-constrained algorithm) against market available devices shows that the average detection error of the developed algorithm is approximately 2% under various test conditions (i.e., walking, fast walking, and running) and has an accuracy of 94.96% for activity detection. This study reveals that the use of the proposed algorithm for activity detection not only brings down the cost of the overall device but also reduces the power consumption (< 45.7 mW) with a minimal covering size (12.56 cm2) for the wearable part that can extend the application area of such devices.