Enhancing Environmental Independency for CSI-based Depth Image Reconstruction

Guanyu Cao, Takuya Maekawa, Kazuya Ohara, Yasue Kishino

Abstract: This study presents a preliminary investigation of enhancing environmental independency in reconstructing depth images of moving objects using Wi-Fi channel state information (CSI). The Wi-Fi-based depth imaging technique has promising application scenarios such as surveillance and elder care. However, CSI is greatly affected by the device’s location and background objects, such as furniture, making it difficult to reconstruct depth images from CSI in new environments unless collecting a large amount of data in the new target environments. To address this, we employ few-shot training data collected in the target environment to aid training of mapping between CSI and depth images. Our method mitigates the labor to collect training data in new environments and improves the model’s robustness.

External IDs:dblp:conf/percom/CaoMOK25