Towards data center stocktaking Using computer vision

This paper presents a computer vision-based approach of an automated stocktaking of servers in a data center using a camera and an inertial measurement unit. During the process, labels of servers are first detected and deprojected in an absolute frame of reference. The distance measure between servers are then computed and analyzed to state whether labels are missings. Experiments were carried out at OVHcloud’s experimental data center, using an iPhone to collect the data. iPhone sensor data was collected with the ARKit library and an average absolute location error of 3.16 mm was estimated for the worst situations.