A Study On Active Camera Relocalization With RGDB Camera

As one fundamental research area in computer vision,graphics and robotics,Ac-tive Camera Relocalization?ACR?aims to dynamically and actively adjust camera pose to make it the same as the one the reference image is taken.So the captured image will be the same as reference image if lighting condition and changes of the scene are ignored.This paper discusses active camera relocalization methods with RGBD camera,with improvements on the existing platform and algorithms^[1],we solve the problem that needs too many iterations and long running time for ACR converge caused by the missing of scale information.Theoretically,we fill the blank of ACR with RGBD camera by providing one framework and discuss various extensions with different data modal and relax the prerequisites at the same time,thus provide more theoretical sup-port for depth camera adoption.Practically,the 6D platform and system can physically relocate camera pose with high precision.By changing previous bisection method for movement with method aided with additional depth information,we increase the con-verging speed noticeably,so that it becomes more convenient to get accurate relocated images,and is beneficial to apply the platform to more scenarios.What is more,the virtual environment built upon UnrealCV^[2]makes it easier to reproduce our work.ACR is still at its initial stage at present,relocate camera pose conveniently and quickly can have great effort on many research areas,such as minute change monitoring,panorama stitching,robotics?e.g.automated picking and stowing?,virtual reality,3D scene reconstruction and etc.This paper brings ACR to one new level and advances the development and application of ACR.