Recent progress in sensing and computer vision has enabled the application of deep learning (DL)-based CV technology in millimeter and terahertz beamforming. In CV-based beamforming, a pre-trained object detector is used to acquire the channel paramet...
Recent progress in sensing and computer vision has enabled the application of deep learning (DL)-based CV technology in millimeter and terahertz beamforming. In CV-based beamforming, a pre-trained object detector is used to acquire the channel parameter information (distance and angle) of the wireless object, using which the BS can generate a directional beam heading toward the wireless object. In this paper, we propose a novel beamforming technique that exploits an object detector tailored for identifying wireless objects to maximize the beamforming gain. To develop such detector, we collected a massive vision dataset called Vision Objects for Millimeter and Terahertz Communications (VOMTC), which consists of RGB and depth images of people holding cell phones and laptops. Through beamforming experiments using the VOMTC test set, we show that the proposed technique outperforms conventional beamforming techniques in terms of data rate.