Optics and Lasers in Engineering 42 (2004) 179–192 Object reconstruction in multilayer neural network based profilometry using grating structure comprising two regions with different spatial periods Dinesh Ganotra, Joby Joseph, Kehar Singh* Photonics Group, Department of Physics, Indian Institute of Technology Delhi, New Delhi 110016, India Received 13 March 2003; accepted 7 August 2003 Abstract Feed-forward backpropagation neural network has been used in fringe projection profilometry for reconstruction of a three-dimensional (3D) object. A grating structure comprising two regions of different spatial periods is projected on the reference surface over which the object is placed. The shorter spatial period part of the grating is projected over the object, whereas the longer spatial period part is projected on the reference surface only. 3D object shape is reconstructed with the help of neural networks using images of the projected grating. During training phase of the network, the shorter spatial period grating along with the longer spatial period grating is used. Experimental results are presented for a diffuse object, showing that the 3D shape of the object is recovered using the above-mentioned method. However, the phases wrapping takes place in Fourier transform profilometry by using only one grating of shorter spatial period. r 2003 Elsevier Ltd. All rights reserved. Keywords: Neural networks; Profilometry; 3D object shape reconstruction 1. Introduction Real-time three-dimensional (3D) measurement is an ongoing effort in industry. Such measurements have applications in control of intelligent robots, obstacle detection for vehicle guidance, dimension measurement for die development, ARTICLE IN PRESS *Corresponding author. Tel.: +91-11-2659-1324; fax: +91-11-2658-1114. E-mail address: kehars@physics.iitd.ernet.in (K. Singh). 0143-8166/$-see front matter r 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.optlaseng.2003.08.002