Fast and Effective Visual Place Recognition using Binary Codes and Disparity Information Roberto Arroyo 1 , Pablo F. Alcantarilla 2 , Luis M. Bergasa 1 , J. Javier Yebes 1 and Sebasti´ an Bronte 1 Abstract—We present a novel approach for place recognition and loop closure detection based on binary codes and disparity information using stereo images. Our method (ABLE-S) applies the Local Difference Binary (LDB) descriptor in a global framework to obtain a robust global image description, which is initially based on intensity and gradient pairwise comparisons. LDB has a higher descriptiveness power than other popular alternatives such as BRIEF, which only relies on intensity. In addition, we integrate disparity information into the binary descriptor (D-LDB). Disparity provides valuable information which decreases the effect of some typical problems in place recognition such as perceptual aliasing. The KITTI Odometry dataset is mainly used to test our approach due to its varied environments, challenging situations and length. Additionally, a loop closure ground-truth is intro- duced in this work for the KITTI Odometry benchmark with the aim of standardizing a robust evaluation methodology for comparing different previous algorithms against our method and for future benchmarking of new proposals. Attending to the presented results, our method allows a fast and more effec- tive visual loop closure detection compared to state-of-the-art algorithms such as FAB-MAP, WI-SURF and BRIEF-Gist. I. I NTRODUCTION Visual navigation systems used by intelligent robots and vehicles need effective and efficient computer vision al- gorithms for robust and fast place recognition. Since the appearance of FAB-MAP [1], loop closure techniques based on visual information have experienced a significant growth and several works have contributed to this research line (see Section II). However, FAB-MAP has some drawbacks, such as the need of a previous training to build a visual vocabulary of the environment and the associated probabilistic approach, making the algorithm not suitable for real-time applications. The usage of binary descriptors for place recognition has been recently proposed in some works in the litera- ture [2], [3], [4], [5]. Binary descriptors provide a reduction in computational costs and memory resources compared to vector-based descriptors. In [2], the authors introduced a global binary descriptor named BRIEF-Gist for loop closure detection problems. BRIEF-Gist was shown to accomplish better detection rates than FAB-MAP, while being several orders of magnitude faster. Besides, BRIEF-Gist also has better performance than global vector-based descriptors such as WI-SURF [6], as demonstrated in our tests. *This work is funded by the UAH through a FPI grant, the Spanish MINECO through the project Smart Driving Applications (TEC2012-37104) and the CAM through the project RoboCity2030 II (S2009/DPI-1559). 1 Department of Electronics, University of Alcal´ a (UAH), Alcal´ a de Henares, 28871, Madrid, Spain. {roberto.arroyo, bergasa, javier.yebes, sebastian.bronte}@depeca.uah.es 2 Toshiba Research Europe Ltd., Cambridge, United Kingdom. pablo.alcantarilla@crl.toshiba.co.uk Fig. 1. Proposed visual place recognition method based on LDB codes and disparity information for loop closure detection. Given an image patch p and a sampling pattern, a binary test is applied as a comparison of the image intensity, gradient and disparity. In this paper, we propose a novel approach for visual place recognition and loop closure detection which uses global binary descriptors that are built from intensity, gradient and disparity pairwise comparisons, as presented in Fig. 1. We extend the Local Difference Binary (LDB) [7] descriptor to incorporate disparity information (D-LDB). As will be shown in our experiments, the addition of disparity provides a more precise visual localization than only using intensity and gra- dient information. The image description applying disparity helps to reduce some typical problems related to visual place recognition, such as perceptual aliasing. Our method, named ABLE-S, improves the results obtained by other state-of-the-art algorithms such as FAB-MAP, WI-SURF and BRIEF-Gist and it also has a low computational cost. The most important contributions introduced in this work are the following: • An innovative algorithm for visual place recognition and loop closure detection based on LDB global binary descriptors and disparity (see Section III and IV). • A new ground-truth designed for loop closure detection in a challenging dataset as the KITTI Odometry [8] and an objective evaluation methodology (see Section V). • An extensive comparative study about the experimental results obtained with our algorithm in its different variants against the principal state-of-the-art algorithms (see Section VI) and the main conclusions obtained jointly with some future research lines (see Section VII). 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) September 14-18, 2014. Chicago, IL, USA, 978-1-4799-6933-3/14/$31.00 ©2014 IEEE 3089