Semi-Automatic Linear Feature Extraction by Dynamic Programming and LSB-Snakes

This paper deals with semi-automatic linear feature extraction from digital images for GIS data capture, where the identification task is pe$ormed manually on a single image, while a special automatic digital module performs the high precision feature tracking in two-dimensional (2-0) image space or even three-dimensional (3-0) object space. A human operator identifies the object from an on-screen display of a digital image, selects the particular class this object belongs to, and provides a very few coarsely distributed seed points. subseq;ently, with th;?sk seed as an approximation of the ~os i t ion and s h a m the linear feature will be extracted automatically by either a dynamic programming approach or by L S B S ~ ~ ~ ~ S [Least-Squares E-spline Snakes). With dynamic programming, the optimization problem is set up as a discrete multistage decision process and is solved b y a "timedelayed" algorithm. It ensures global optimality, i s numerically stable, and allows for hard constraints to be enforced on the solution. In the least-squares approach, we combine three types of observation equations, one radiometric, formulating the matching of a generic object model with image data, and two that express the internal geometric constraints of a curve and the location of operator-given seed points. The solution is obtained by solving a pair of independent normal equations to estimate the parameters of the spline curve. Both techniques can be used in a monoplotting mode, which combines one image with its underlying DTM. The L S B S ~ ~ ~ ~ S approach is also implemented in a multi-image mode, which uses multiple images simultaneously and provides for a robust and mathematically sound full 3 D approach. These techniques are not restricted to aerial images. They can be applied to satellite and close-range images as well. The issues related to the mathematical modeling of the proposed methods are discussed and experimental results are shown in this paper too. Introduction One of the most fascinating promises of digital photogrammetry is the highly automated acquisition and updating of spatial data from images. Remarkable progress has been made in areas involving image and template matching such as automatic interior orientation, relative orientation, tie point selection and transfer, digital terrain model (DTM) generation, and digital ortho-image generation. Although the current level of automation on most digital photogrammetric stations is still fairly low, a number of these developments are meanwhile available on some commercial systems (Gruen, 1996; Miller et al., 1996; Walker and Petrie, 1996). On the way towards automatic mapping or spatial data acquisition and update, automatic identification and localization of cartographic objects in aerial and satellite images has Institute of Geodesy and Photogrammetry, Swiss Federal Institute of Technology Zurich, ETH-Hoenggerberg, CH-8093 Zurich, Switzerland ([email protected]). PE&RS August 1997 gained increasing attention in recent years in digital photogrammetry and computer vision. Despite the fact that quite some achievements have been reported, the automatic extraction of man-made objects in essence is still an unresolved issue. As fully automatic methods for mapping are still far out of reach, semi-automatic methods for feature extraction that interact with a human operator are considered to be a good compromise, combining the mensuration speed and accuracy of a computer algorithm with the interpretation skills of a human operator. Although the level of automation is much higher in close-range applications, many approaches there are restricted to the measurement of point-shaped features. Automated linear feature extraction has been reported only occasionally (e.g., Gruen and Stallmann, 1991; Streilein, 1996). We have developed two semi-automated algorithms for linear feature extraction, to be used on any type of image (satellite, aerial, close-range). We have treated with these techniques images of topographical, architectural, industrial, and medical objects, and there are no restrictions concerning applications. Linear edge-type and ribbon-like features may be extracted. The techniques of choice are based cm a method of dynamic programming and on the estimation of energy-minimizing functions ("Snakes"). We have developed a new kind of Snake class, the LSB-snakes (Least Squares Bspline Snakes), which combine the powerful tools of leastsquares estimation with the determination of energy-minimizing functions. This article introduces our general semi-automatic linear feature extraction scheme. For both techniques, it summarizes the respective mathematical approaches. We focus here in particular on the extraction of roads and we emphasize the underlying generic road models. Monoplotting and multiple-image implementations will be shown. Due to space limitations not all details of the algorithms can be presented here. For more information, please refer to the relevant publications (Gruen and Li, 1995; Gruen and Li, 1996; Li, 1997). General Semi-Automatic Feature Extraction Scheme Our semi-automatic feature extraction scheme is shown in Figure 1. In the current implementation, the main procedures of image preprocessing are Wallis filtering and road sharpening. The Wallis filter is used to enhance the images and facilitate the subsequent road extraction process by locally forcing the grey value mean and especially contrast (dynamic range) to fit certain target values (Baltsavias, 1991). This preprocessing stage is particularly important for panchromatic Photogrammetric Engineering & Remote Sensing, Vol. 63, NO. 8, August 1997, pp. 985-995. 0099-11l2/97/6308-9&5$3.00/0