Int J Parallel Prog (2015) 43:703–720 DOI 10.1007/s10766-014-0316-7 A Comparative Study of Parallel RANSAC Implementations in 3D Space Alejandro Hidalgo-Paniagua · Miguel A. Vega-Rodríguez · Nieves Pavón · Joaquín Ferruz Received: 1 October 2013 / Accepted: 27 June 2014 / Published online: 10 July 2014 © Springer Science+Business Media New York 2014 Abstract RANSAC (RAndom SAmple Consensus) is an iterative method for estimating the parameters of a certain mathematical model from a set of data which may contain a large number of outliers (noisy points). The main problem of the RANSAC algorithm is that it is too expensive in terms of execution time when real-time processing is needed (30 fps). In view of the importance of this algorithm and the rise of parallelism-based technologies, we analyze and compare in this work various parallel implementations based on different techniques (OpenMP, POSIX Threads, and CUDA). To the best of our knowledge, no other articles have attempted a similar study. In order to make this first study, we have used some standard metrics in parallelism (Runtime and Speedup) and some specific metrics used in evaluating search strategies (Precision, Recall, and F-Score). Furthermore, the experiments have been executed in different hardware alternatives in order to present a more complete study. The conclusions of our study show the advantages and disadvantages of the different parallel implementations. A. Hidalgo-Paniagua (B ) · M. A. Vega-Rodríguez Department of Technologies of Computers and Communications, Polytechnic School, University of Extremadura, Cáceres, Spain e-mail: ahidalgop@unex.es M. A. Vega-Rodríguez e-mail: mavega@unex.es N. Pavón Department of Information Technology, Higher Technical School of Engineering, University of Huelva, Huelva, Spain e-mail: npavon@dti.uhu.es J. Ferruz Department of Systems Engineering and Automation, Higher Technical School of Engineering, University of Sevilla, Sevilla, Spain e-mail: ferruz@cartuja.us.es 123