To be presented at the third Asian Conference on Computer Vision, Hong Kong, 8-11 January, 1998. Robust Motion Segmentation Using Rank Ordering Estimators Alireza Bab-Hadiashar and David Suter Intelligent Robotics Research Centre, Department of Electrical & Computer Systems Engineering, Monash University, Clayton Vic. 3168, AUSTRALIA. E-mail: [ali, suter]@basil.eng.monash.edu.au Abstract Robust estimators have become popular tools for solving a wide range of problems in computer vision. Despite many successes in this field, there is still a need for estimators, which are suited to specific problems such as recovering structures from multi-structural data. This paper offers an alternative approach to, and some practical insights into, the implementation of well-known rank ordering based robust estimators. The approach has been tested on synthetic and real image data for motion segmentation purposes. 1 Introduction This paper describes an alternative approach to, and some practical insights into, the implementation of well-known rank ordering based robust estimators for motion segmentation purposes. These rank ordering based estimations are of interest because of their high break down points. High breakdown point estimators are frequently employed to solve various computer vision problems. Although some of the estimators were developed within computer vision (Stewart, 1997), almost all of them find their roots in the field of robust statistics. Indeed, the notion of fitting a parameterized model to a set of noisy data has been considered by statisticians for many years. Several robust methods have been proposed to solve many problems in different scientific fields. However, there are still problems, particularly in computer vision, for which a satisfactory method is yet to be found. This is mainly because many problems in this field are inherently complicated as they involve fitting models to multi- structural data (a set of data, which can be partitioned to different groups, each group fitting to different parametric models). In this paper, a new approach to the general problem of fitting parametric models to multi-structural data is considered. A robust estimator named Selective Statistical Estimator (SSE) is developed. The SSE is based on using the least K-th Order squares (LKS) with some intuition employed to resolve the issue of what would be the best value for K. The rest of this paper is organized as follows. The recent history of rank ordering statistics and its usage in computer vision is discussed in section 2. Section 3 offers a brief discussion on the computational cost of some of the existing robust rank ordering based estimators. The fundamental notion behind the new approach is explained in section 4. Experimental results are presented in section 5 and section 6 concludes the paper. 2 Robust Estimation The essential idea to be developed in this paper is that too many researchers are overly concerned with retrieving the major population first. To see why this may have