Acoustic imaging in the rayspace : application to environment inference

This work is part of the environment-aware processing, a field of recent interest that promises to significantly push the boundaries of audio signal processing. The environment-aware processing uses information arising from the environmental response to enable the acoustic systems to become aware of their own characteristics and geometry and those of the environment that they operate in. This information allows advanced and innovative space-time processing solutions. In particular the thesis addresses the problem of inference from acoustic measurements on the geometric characteristics of the environment. Recently a number of techniques for localization of reflective surfaces appeared in literature. These techniques exploit different information extracted from acoustic measurements to infer the position of the reflectors in the environment. Usually the extracted information, combined with some a priori knowledge, defines a non-linear constraint on reflector position. Using multiple constraints (e.g changing the hardware position) a cost function is formulated whose minimization yields the estimated line or plane (for 2D or 3D geometries) on which the reflector lies. In this work we take a slightly different approach for the localization of reflective surfaces. Instead of extracting information related to a specific geometric constraint, we are interested in "looking" at the acoustic scene, i.e. obtaining an overview of what is happening in different positions in space, and successively estimating the environment geometry from a number of such acoustic "snapshots". Therefore, we want to imitate, to a certain extent, the procedures used in computer vision to reconstruct the environment geometry taking visual snapshots from different points of view. The acoustical snapshots are defined using a non-linear transformation applied to acoustic measurements that maps the data in a space in which the geometric primitives are represented by linear constraints. Unlike most of other methods, the acoustic observation of the environment allows us to find not only the line on which the reflector lies but also its extension. This property can turn useful in irregular, complex environments where occlusions and limited visibility of acoustic reflectors are present. Furthermore, the representation of acoustic measurements defined in this work can potentially be used also to infer on radiometric properties of the environment (e.g. radiation pattern and reflection coefficients) and therefore it has a number of potential other applications.