TOWARD A ONE SHOT MULTI-PROJECTOR PROFILOMETRY SYSTEM FOR FULL FIELD OF VIEW OBJECT MEASUREMENT Stuart Woolford, Ian. S. Burnett School of Electronic and Computer Engineering, RMIT University ABSTRACT In this paper a one-shot method to determine the shape of an object from overlapping cosine fringes projected from multiple projectors is presented. This overcomes the limitation with single projector systems that do not allow imaging the entire object with a single shot. The proposed method projects orthogonal fringe patterns from different projectors and uses Fourier domain filtering to isolate the fringes, which are demodulated using an unscented particle filter. Sources of error are discussed and their effects on the resulting parameter estimation are shown, as well as methods to reduce their impact. The proposed method is tested on simulations and real world objects and it is shown to be effective to isolate interfering fringes and determine the shape of an object. Index Terms— Multi-View Profilometry, Structured Light Projection, Unscented Particle Filter, Multiple Projector Structured Light 1. INTRODUCTION Most multi-view techniques have so far focused on imaging single view at a time and connecting the images together. Zheng, Guo et al. [1] used an N step phase shifting method to first image the object, and then employed a multi-view connection technique using quaternion based coordinate transform and multi-aperture overlap scanning technique (MAOST) to connect the views. A similar connection technique was used in [2] where the point cloud of the measured object is transformed into cylindrical coordinate system and connected via MAOST. Virtual cylinders were used in [3] to overcome the problem with complex shapes not being represented correctly with cylindrical transforms. Multiple projector methods were used in [4, 5] for full field of view shape measurement. In [4], parallel colour de-Bruijn patterns projected from multiple projectors was used for full object detection, while a similar method was used in [5]. However the method of Furukawa et al. requires overlap of each pattern to form a grid in order to process the patterns. Gai and Su [6] presented a multiple projector profilometry system based on inverse function analysis but this requires knowledge of the shape in question and can’t be done blind. A colour scheme was presented in [7, 8] where fringes of different colours are projected from each projector. Bayesian methods have been employed in the past to determine object shape via optical profilometry. Villa, Servin and Castillo [9] employed shape measurement using regularized filters with a Markov random field as a prior. Recursive Bayesian estimation has been attempted before in interference fringe analysis [10-13]. Gurov, Ermolaeva and Zakhrov [10] used a Kalman filter to analyse low-coherence fringes in an interferometry system, however this method still required phase unwrapping, and was only tested on surfaces with up to 100μm. The unscented Kalman Fig. 1. Setup For The Multi-Projector Profilometry System filter was used for phase-step interferometry in [14], using a model to estimate the phase and amplitude of an interferometric fringe. This method required phase unwrapping and was only tested for small deformations. This paper relates to the multi-projector methods in [4, 5], in that each pattern projected from each projector is at right angles to the others. However this paper presents a method where each pattern can be processed individually while allowing for overlap of adjacent patterns thereby reducing the number of projectors required to image the entire object, although it is not fully automatic as the pattern must be selected manually. It also expands on the multi-view one-shot profilometry techniques as opposed to multi-view methods that employ N-ary phase shifting profilometry, which require N shots per view. The proposed method expands on the non-linear filtering methods such as [10-13], however uses an Unscented Particle Filter, which is less sensitive to pixel offset of the input fringe than the standard UKF and increases the estimation accuracy over Bayesian filters by employing the UKF as a proposal distribution. 2. MULTIVIEW PROILOMERTY SETUP Figure 1 shows the base system setup of the multi-projector profilometry system. Projector-1 (  ) and Projector-2 (  ) both project onto the object at the same time, with projection field for    between α r and α l , and projection field for    between β r and β l . In order to capture the full field of view for the object the fringe patterns must overlap crossing at point C. In order to capture the height of the object the fringe pattern from a single projector is projected onto a virtual reference plane to point D. The reference plane is then removed and the fringe pattern is projected 2014 IEEE International Conference on Acoustic, Speech and Signal Processing (ICASSP) 978-1-4799-2893-4/14/$31.00 ©2014 IEEE 569