Elsevier

Image and Vision Computing

Volume 23, Issue 8, 1 August 2005, Pages 707-720
Image and Vision Computing

Optimised De Bruijn patterns for one-shot shape acquisition

https://doi.org/10.1016/j.imavis.2005.05.007Get rights and content

Abstract

Coded structured light is an optical technique based on active stereovision which allows shape acquisition. By projecting a suitable set of light patterns onto the surface of an object and capturing images with a camera, a large number of correspondences can be found and 3D points can be reconstructed by means of triangulation. One-shot techniques are based on projecting an unique pattern so that moving objects can be measured. A major group of techniques in this field define coloured multi-slit or stripe patterns in order to obtain dense reconstructions. The former type of patterns is suitable for locating intensity peaks in the image while the latter is aimed to locate edges. In this paper, we present a new way to design coloured stripe patterns so that both intensity peaks and edges can be located without loss of accuracy and reducing the number of hue levels included in the pattern. The results obtained by the new pattern are quantitatively and qualitatively compared to similar techniques. These results also contribute to a comparison between the peak-based and edge-based reconstruction strategies.

Introduction

The shape acquisition problem has long been a challenge for the computer vision community [1], [2]. The effort put into finding solutions has led to a large taxonomy of non-contact optical techniques [3]. In this field, two major groups of techniques can be distinguished: passive and active. Passive techniques are based on obtaining the shape of the object without introducing energy into the scene. Some of the cues used are shape from motion, texture and shadows which are extracted by using a single image or a set of images captured by a single camera or multiple cameras. One of the most well-known passive techniques is stereovision [4]. The main difficulty of stereovision consists of solving the correspondence problem between the different views of the same object. Even if the correspondence problem can be made easier thanks to geometrical constraints [5], its solution is still difficult and computationally expensive. Furthermore, stereovision systems cannot ensure a maximum resolution; in fact, the number of correspondences and the number of 3D points is quite reduced because it is highly dependent on the observed object. Moreover, when the measured object does not contain singular points, such as corners, or it is not highly textured, correspondences cannot be found. Despite these limitations, stereovision techniques still have an outstanding potential thanks to the huge amount of information that can be obtained by using cameras. Investigations into how to get the most from this potential led to a powerful solution: active stereovision. In general terms, active optical techniques project light onto the object and then the reflected light is measured; in this way, information on shape and range can be calculated. In the specific case of active stereovision, controlled illumination is projected onto the object and then a camera, or a set of cameras, is used to capture the images. The controlled illumination is used to project singular points onto the object surface. Therefore, the correspondence problem can be solved easily.

In the development of active stereovision techniques, structured light was the first to appear. Structured light techniques basically involve projecting laser planes or laser dots and scanning the object surface. An image must be captured for every position of the scanning laser. For each image, all the pixels which are illuminated by the laser can be triangulated and reconstructed. The main advantages of structured light techniques include the easy image processing involved and the high accuracy that can be achieved in the 3D reconstruction. In some of these techniques, however, the mechanical operations required or the number of images that must be taken can become a drawback.

Coded structured light can be considered as a logical evolution of classical structured light techniques. The aim of this group of techniques is to increase the resolution achieved in every image. Such techniques are based on projecting bidimensional light patterns by using devices such as slide projectors and, more recently, LCD (liquid-crystal device) and DMD (digital mirror device) projectors. The systems based on these devices are more flexible since the projected patterns can be easily and inexpensively changed. The patterns are designed so that a set of points are encoded, i.e. its pattern position can be retrieved from the camera image(s) by gathering local information around this encoded point. A key element of coded structured light techniques is the coding strategy used to generate the light patterns. A coding strategy is needed to avoid ambiguities in the projected patterns, i.e. different pattern regions being equal and therefore, indistinguishable. The coding strategy used strongly determines the performance of the technique in terms of resolution, accuracy, and the number of required patterns.

Two main groups of techniques based on coded structured light exist [6]. The first group, called time-multiplexing, is based on projecting a sequence of patterns so that colour is not required but only static objects can be measured. The second group is based on projecting a unique pattern and it is referred as one-shot techniques. In this paper we focus on one-shot techniques based on a coloured pattern. Concretely, the paper presents a new way of encoding coloured one-shot patterns. Even if this type of patterns is usually limited to colour neutral or pale objects, they can obtain very good results in terms of resolution and accuracy. The aim of the new coding strategy is to define a hybrid pattern so that it improves the performance of similar existing patterns in terms of resolution and accuracy. The structure of the paper is as follows. First, in Section 2, a brief overview of one-shot techniques based on coded structured light is presented, focusing on techniques using colour. Secondly, in Section 3, we describe two of the most usual coloured patterns in one-shot techniques and we present the new pattern which combines the advantages of both of them. Afterwards, details about the segmentation and decoding processes for the new pattern are exposed in Section 4. The system's calibration is briefly explained in Section 5. Then, the hybrid pattern is compared to similar patterns through experimental results from a quantitative and a qualitative point of view in Section 6. Finally, the conclusions of the paper are discussed in Section 7.

Section snippets

Overview of one-shot techniques

According to an early study of the state of the art on coded structured light [6], it can be stated that most part of one-shot techniques are based on spatial coding. In time-multiplexing techniques the labels of the encoded points are multiplexed along a sequence of patterns. In spatial coding (one-shot techniques) the label is encoded in a local neighbourhood. Therefore, while time-multiplexing techniques assume that the object remains static, spatial techniques generally assume that the

A proposal of a new hybrid pattern

In this section, we present a brief review of two classical one-shot techniques and afterwards a new hybrid pattern combining the advantages of both of them.

As already mentioned in Section 2, there is an important group of one-shot techniques which are based on coloured multi-slit and stripe patterns. Multi-slit patterns introduce black gaps between the coloured bands of pixels so that two consecutive slits can have the same colour. The black gaps also allow intensity peaks to be detected in

Pattern segmentation and decoding

Given the system configuration of our experimental setup, i.e. a camera and a projector positioned aside and the pattern consisting of vertical stripes, the decoding process is based on horizontal scan-lines. In case that the projector and the camera are not approximately aside, a stereopair rectification algorithm can be used to transform the images taking into account the geometry of the system [27].

According to the aim of the proposed hybrid pattern, for every scan-line, the intensity peaks

System's calibration

In this section, we present the different modelling steps that must be considered in order to obtain the 3D reconstruction of the illuminated scene. First of all, we explain how the geometric calibration of the system is performed. Afterwards, some imperfections of the real system are modelled in order to increase the robustness and the accuracy.

Experimental results

In this section, we show some experimental results validating the proposed hybrid pattern. The experimental setup consists of an Olympus Camedia digital camera and a Mitsubishi XL1U LCD projector which are positioned aside with a relative direction angle of about 15°. Both devices operate at 1024×768 pixels. The measuring volume is about 30 cm high, 40 cm wide, and 20 cm deep.

The calibration of the system has been performed according to the steps presented in Section 5 and summarised in Fig. 5. The

Conclusions

This paper presents a new coloured pattern for one-shot shape acquisition. The new pattern aims to improve similar existing patterns in terms of resolution and accuracy. Among the one-shot colour-based techniques two of the most frequent patterns are based on multi-slits or stripes. The former is based on coloured bands of pixels (slits) separated by black gaps so that consecutive bands can have the same hue value. In the latter, the coloured bands are adjacent (stripes) so that two consecutive

Acknowledgements

Work partially founded by the Ministry of Universities, Research and Information Society of the Catalan Government.

References (32)

  • R. Jarvis

    Range sensing for computer vision

    (1993)
  • F. Chen et al.

    Overview of three-dimensional shape measurement using optical methods

    Optical Engineering

    (2000)
  • O. Faugeras

    Three-Dimensional Computer Vision: A Geometric Viewpoint

    (1993)
  • T. Etzion

    Constructions for perfect maps and pseudorandom arrays

    IEEE Transactions on Information Theory

    (1988)
  • C.J. Davies et al.

    A hough transform for detecting the location and orientation of 3-dimensional surfaces via color encoded spots

    IEEE Transactions on Systems, Man and Cybernetics

    (1998)
  • H.J.W. Spoelder et al.

    Some aspects of pseudo random binary array-based surface characterization

    IEEE Transactions on Instrumentation and Measurement

    (2000)
  • Cited by (0)

    View full text