Using Smart Insoles and RGB Camera for Identifying Stationary Human Targets

Identity recognition is an important component for creating a personalized service in loT applications. Current prevailing technologies have to pre-train with large datasets or need the privacy-sensitive information from users such as facial features, voice features, and fingerprint. In this work, we address the problem of identifying stationary humans (less movements) targets, which cannot be solved by other motion-based fusion mechanisms. In the future IoT world, many wearable sensors on human beings are foreseeable. We exploit RGB camera and smart insoles to design a system for dealing with the stationary identity recognition. We utilize machine learning algorithms to explore the correlation of lower body postures from the viewpoints of heterogeneous sensors. Then we can make the identity matching according to the trained models. Evaluation results show that our mechanism can achieve good performance if users' postures are differentiable.