2D–3D spatial registration for remote inspection of power substations

Abstract

Remote inspection and supervisory control are critical features for smart factories, civilian surveillance, power systems, among other domains. For reducing the time to make decisions, operators must have both a high situation awareness, implying a considerable amount of data to be presented, and minimal sensory load. Recent research suggests the adoption of computer vision techniques for automatic inspection, as well as virtual reality (VR) as an alternative to traditional SCADA interfaces. Nevertheless, although VR may provide a good representation of a substation’s state, it lacks some real-time information, available from online field cameras and microphones. Since these two sources of information (VR and field information) are not integrated into one single solution, we miss the opportunity of using VR as a SCADA-aware remote inspection tool, during operation and disaster-response routines. This work discusses a method to augment virtual environments of power substations with field images, enabling operators to promptly see a virtual representation of the inspected area's surroundings. The resulting environment is integrated with an image-based state inference machine, continuously checking the inferred states against the ones reported by the SCADA database. Whenever a discrepancy is found, an alarm is triggered and the virtual camera can be immediately teleported to the affected region, speeding up system reestablishment. The solution is based on a client-server architecture and allows multiple cameras deployed in multiple substations. Our results concern the quality of the 2D–3D registration and the rendering framerate for a simple scenario. The collected quantitative metrics suggest good camera pose estimations and registrations, as well as an arguably optimal rendering framerate for substations' equipment inspection.