A wearable-based posture recognition system with AI-assisted approach for healthcare IoT

Abstract Human posture recognition is a challenging task in the medical healthcare industry, when pursuing intelligence, accuracy, security, privacy, and efficiency, etc. Currently, the main posture recognition methods are captured-behaviors-based visual image analysis and wearable devices-based signal analysis. However, these methods suffer from issues such as high misjudgment rate, high-cost and low-efficiency. To address these issues, we propose a collaborative AI-IoT-based solution (namely, WMHPR) that embeds with advanced AI-assisted approach. In WMHPR, we propose the multi-posture recognition (MPR), an offline algorithm is implemented on wearable hardware, to identify posture based on multi-dimensions data. Meanwhile, an AI-based algorithm running on the cloud server (online), named Cascade-AdaBoosting-CART (CACT), is proposed to further enhance the reliability and accuracy of MPR. We recruit 20 volunteers for real-life experiments to evaluate the effectiveness, and the results show our solution is significantly outstanding in terms of accuracy and reliability while comparing with other typical algorithms.