Rapidly developing computer vision algorithms bring us an opportunity that a machine improves our lives by analyzing the environment autonomously (e.g., Amazon Go). However, because of the vision’s inherent limitations such as occlusion, fragility to the dark, and privacy invasion, there are places where it is difficult to introduce a vision system (e.g., toilet, indoor without light). This paper presents the first battery-free near-infrared (NIR) camera tag system which enables deep learning based vision services. Our system solves three important problems in applying the dazzlingly developing computer vision algorithms to our lives; (i) Privacy Invasion, (ii) Occlusion, and (iii) Unavailability in Darkness. We deal with these problems by designing (i) Extremely low-resolution (eLR) video, (ii) Battery-free, and (iii) IR camera system. We experimented with prototype hardware with COTS elements to verify its feasibility. Furthermore, we confirmed that our system achieves 5 frames per second (FPS) with the tag and reader 5 m apart, and it works even when the tag and light source are 3 m apart. Finally, we discuss that photo safety has been confirmed for possible risks in power transmission using near-infrared rays, and that it can be extended to various services not only indoors but also outdoors (e.g., fall detection, wildfire monitoring).