Wi-Learner

Contactless RF-based sensing techniques are emerging as a viable means for building gesture recognition systems. While promising, existing solutions have poor generalization ability when targeting new users, environments or device deployment. They also often require multiple pairs of transceivers and large number training samples each target domain. These limitations either lead to cross-domain performance incur huge labor cost, hindering their practical adoption. This paper introduces Wi-Learner, novel solution that relies on just one pair but can deliver accurate using data sample per user, environment setup. Wi-Learner achieves this by first capturing the gesture-induced Doppler frequency shift (DFS) from noisy measurements carefully designed signal processing schemes. It then employs convolution neural network-based autoencoder extract low-dimensional features be fed into downstream model recognition. meta-learner "teach" network learn effectively small set points, allowing base quickly adapt domain sample. By so doing, we reduce overhead collection allow system change deployed environment. We evaluate applying it Widar 3.0 dataset. Extensive experiments demonstrate is highly efficient has good ability, delivering an accuracy 93.2% 74.2% - 94.9% in-domain gesture, respectively.