Automatic 3D scanning surface generation for microphone array acoustic imaging Mathew Legg ⇑ , Stuart Bradley Physics Department, University of Auckland, Auckland, New Zealand article info Article history: Received 13 December 2012 Received in revised form 12 June 2013 Accepted 12 August 2013 Available online 14 September 2013 Keywords: Microphone array 3D Acoustic imaging Structured light Beamforming CLEAN-SC abstract This work presents a new technique for automatically generating the 3D scanning surface for acoustic imaging using microphone arrays. Acoustic images, or maps, of sound coming from spatially distributed sources, may be generated from microphone array data using algorithms such as beamforming. Tradi- tional 2D acoustic maps can contain errors in the near-field if the object being imaged has a 3D shape. It has been shown that using the 3D surface geometry of an object as a scanning surface for beamforming can provide more accurate results. The methods used previously to generate this 3D scanning surface have either required existing CAD (Computer-Aided Design) models of the object being acoustically imaged or have required separate equipment which is generally bulky and expensive. The new method uses one or more cameras in the array, a data projector, and structured light code to automatically gen- erate the 3D scanning surface. This has the advantage that it is inexpensive, can be incorporated as an add-onto existing microphone arrays, has short scan time, and is capable of being extended to imaging dynamic scenes. This technique is tested using beamforming and CLEAN-SC (CLEAN based on spatial Source Coherence) algorithms for a spherical array and an Underbrink multi-arm spiral array. For sound sources located about 1.2 m from the array, the mean position errors obtained are 6 mm. This is a quarter of the diameter of the mini-speakers being used as a sound sources. Ó 2013 Elsevier Ltd. All rights reserved. 1. Introduction The microphone phased array (also known as the acoustic cam- era) has been developed as a tool to enable the position and mag- nitude of sound sources to be identified. Microphone phased arrays are widely used by industries such as aeroplane and automotive manufacturers to identify sound sources. A commonly used acous- tic imaging algorithm is beamforming [1,2], which uses delaying and summing of microphone channel data to obtain acoustic images or ‘maps’. However, the beamforming maps contain an interference pattern artefact referred to as sidelobes. Image sharp- ening or deconvolution techniques have been developed to remove these sidelobes and attempt to obtain the true sound source distri- bution. Examples of these algorithms are DAMAS (Deconvolution Approach for the Mapping of Acoustic Sources) [3] and CLEAN-SC (CLEAN based on Source Coherence) [4]. The other methods used to generate acoustic maps from microphone phased array data are acoustic holography [5] and inverse methods [6]. 1.1. Acoustic imaging using traditional 2D scanning surfaces Acoustic imaging techniques, such as beamforming and decon- volution, have traditionally used the assumption that the acoustic sources lie on a plane. A 2D scan surface is used which is oriented perpendicular to the array’s principle forward direction (the array Z-axis). This can lead to errors in the resulting acoustic maps if the sound sources are offset from the 2D scanning surface. These errors appear as projection/parallax errors in the plotting of the acoustic maps [7] and incorrectly estimated sound pressure levels (SPL) and location of sound sources [8–10]. These beamforming magnitude and position errors result from incorrect focus (time delays) being used for the beamforming. 1.2. Acoustic imaging using 3D scanning surfaces Beamforming and deconvolution have been performed using a 3D grid [11–14]. However, unless the microphone arrays surround the object being imaged, there is poor resolution in the array Z- axis. Another problem is that these 3D grids can contain a large number of scan points, making deconvolution of these 3D grid beamforming maps very computationally expensive. An alternative technique that has been developed is to use a scanning surface for beamforming which corresponds to the 3D 0003-682X/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.apacoust.2013.08.008 ⇑ Corresponding author. E-mail addresses: m.legg@auckland.ac.nz, mleg010@auckland.ac.nz (M. Legg), s.bradley@auckland.ac.nz (S. Bradley). Applied Acoustics 76 (2014) 230–237 Contents lists available at ScienceDirect Applied Acoustics journal homepage: www.elsevier.com/locate/apacoust