Device Free Human Gesture Recognition with Incremental Learning

Abstract

Human gesture recognition based on device-free Wi-Fi wireless signals has received much attention due to its broad application scenarios and far-reaching development prospects. This technology uses wireless signals from Wi-Fi devices to achieve human pose recognition without capturing optical images and specialized devices. Most of the current recognition methods are based on deep learning. The method has shown good performance, but a tricky problem arises immediately. Dynamic updates for recognition tasks in real-world applications require increasing the variety of recognizable gestures. At the same time, previous data is often no longer available, and updates of deep learning models require a high cost. In this paper, an incremental learning-based framework is proposed. The framework enables the model to gain recognition of new gesture without previous access to previous training data. The superiority of the proposed incremental learning framework can be demonstrated by comparing different experiments.