Acoustic Room Modelling Using 360 Stereo Cameras

In this paper we propose a pipeline for estimating acoustic 3D room structure with geometry and attribute prediction using spherical 360 $^{\circ }$ cameras. Instead of setting microphone arrays loudspeakers to measure parameters specific rooms, simple practical single-shot capture the scene stereo pair cameras can be used simulate those parameters. We assume that objects represented as cuboids aligned main axes coordinate (Manhattan world). The is captured off-the-shelf consumer A cuboid-based model estimated by correspondence matching between images semantic labelling convolutional neural network (SegNet). produce frequency-dependent predictions scene. This is, our knowledge, first attempt in literature use visual estimation object classification algorithms predict properties. Results are compared measurements through calculated reverberant spatial audio reverberation reproduction customized given loudspeaker set up.