Monitoring Social Distancing With Single Image Depth Estimation

The recent pandemic emergency raised many challenges regarding the countermeasures aimed at containing virus spread, and constraining minimum distance between people resulted in one of most effective strategies. Thus, implementation autonomous systems capable monitoring so-called social distance gained much interest. In this paper, we aim to address task leveraging a single RGB frame without additional depth sensors. contrast existing single-image alternatives failing when ground localization is not available, rely on image estimation perceive 3D structure observed scene estimate people. During setup phase, straightforward calibration procedure, scale-aware SLAM algorithm available even consumer smartphones, allows us scale ambiguity affecting estimation. We validate our approach through indoor outdoor images employing calibrated LiDAR + camera asset. Experimental results highlight that proposal enables sufficiently reliable inter-personal monitor social distancing effectively. This fact confirms despite its intrinsic ambiguity, if appropriately driven can be viable alternative other perception techniques, more expensive always feasible practical applications. Our evaluation also highlights framework run reasonably fast comparably competitors, pure CPU systems. Moreover, deployment low-power around corner.