Dynamic picking system for 3D seismic data: Design and evaluation

Abstract

In the framework of data interpretation for petroleum exploration, this paper contributes two contributions for visual exploration aiming to manually segment surfaces embedded in volumetric data. Resulting from a user-centered design approach, the first contribution, dynamic picking, is a new method of viewing slices dedicated to surface tracking, i.e. fault-picking, from 3D large seismic data sets. The proposed method establishes a new paradigm of interaction breaking with the conventional 2D slices method usually used by geoscientists. Based on the 2D+time visualization method, dynamic picking facilitates localizing of faults by taking advantage of the intrinsic ability of the human visual system to detect dynamic changes in textured data. The second, projective slice, is a focus+context visualization technique that offers the advantage of facilitating the anticipation of upcoming slices over the sloping 3D surface. From the reported experimental results, dynamic picking leads to a good compromise between fitting precision and completeness of picking while the projective slice significantly reduces the amount of workload for an equivalent level of precision.

Introduction

Today, surveys based on the interpretation of 3D data are essential for petroleum exploration, modern medicine, archaeology, mechanical engineering, etc. In petroleum and gas exploration, how and where to locate the drilling well is strongly dependent on the relevance of the 3D geological model of the subsurface established by the geologist. To plan drilling, geoscientists design a geological model based on the extent and the distances between inner 3D geo-objects such as geological layers, or other important structures related to tectonic Earth activities. Thus, the potential of the 3D object interaction can be fully exploited to support the decision to locate oil or gas reservoirs.

3D object interaction is a large research area for which numerous works exist concerning issues such as occlusion management and object selection. A taxonomy of 3D occlusion management techniques is given by Elmqvist and Tsigas (2007) to solve depth perception issues for the 3D visualization context. Some approaches such as importance-driven volume rendering can be used to maximize the information embedded in the final image by using an automatic hierarchical 3D object management in terms of importance and sparseness (Viola et al., 2004). All these techniques are widely used in medical applications such as surgical planning (Krüger et al., 2005).

However, in the geological context, a key issue is to precisely construct 3D geo-objects leading to the final geological model. Most often, the segmentation of geological volumetric data, called seismic data, arises as a prerequisite to 3D geo-object interaction. In the computer graphics domain, several volume-rendering techniques are used to directly segment particular objects from 3D data. Most common approaches are based on iso-surface extraction, such as the marching cubes algorithm (Lorensen and Cline, 1987), or pixel-oriented strategies, such as the ray casting algorithm (Drebin et al., 1988). Unfortunately, these approaches rely heavily on the existence of a one-to-one relationship between an object and data intensity. In the seismic imaging framework, the nature of the data that are heavily textured, noisy and characterised by high frequencies prevents using such techniques. Inner objects are not directly available and specific techniques to segment them from the volumetric data have to be developed. In this way, supervised approaches based on exploring volumetric data to manually segment objects are largely requested by experts. Moreover, semi-supervised approaches can be considered. Sketch-based methods address this issue (Owada et al., 2005, Owada et al., 2008).

This paper focuses on the issue of volumetric data interaction concerning manual segmentation. Manual segmentation refers to the perceptuo-motor process whereby a human expert localizes, segments and labels each structure directly on the data set. In the seismic context, this task, called picking, is carried out by selecting points of interest with a pointing device on several slices of a seismic block. Most often, the picking task is an iterative process repeating point selection over slices until the finest description possible of the structure is obtained. Geoscientists currently use a static picking technique consisting of a discrete “slice-by-slice” approach. While manual segmentation seems to be accurate, especially for low signal-to-noise ratios, static picking is extremely demanding in terms of time and workload. Thus we propose modifying the manual segmentation paradigm in order to derive a more efficient interactive process.

Our contribution aims to provide a new interaction method called dynamic picking, which constitutes an alternative to conventional strategies based on 2D visualization or pure volume fly-though solutions. Dynamic picking improves shape understanding using a 3D view with animations while reducing the segmentation time and keeping the ease of use of a 2D selection. Moreover, in order to provide more global information on the targeted geological structure while maintaining the integrity of the data, we introduce projective slice as a new focus+context visualization technique.

In the following section, the related works and the geological background from which this work was developed are introduced. Conventional static picking used for manual segmentation in a seismic framework is discussed. Next, the new paradigm of dynamic picking and its advantages are described through an experimental comparison with static picking. Finally, the projective slice is presented and an empirical assessment shows its advantages.