Stereo obstacle detection in challenging environments: the VIAC experience Alberto Broggi, Michele Buzzoni, Mirko Felisa and Paolo Zani VisLab – Dipartimento di Ingegneria dell’Informazione Universit` a degli Studi di Parma, ITALY http://www.vislab.it {broggi,buzzoni,felisa,zani}@vislab.it Abstract— Obstacle detection by means of stereo-vision is a fundamental task in computer vision, which has spurred a lot of research over the years, especially in the field of vehicular robotics. The information provided by this class of algorithms is used both in driving assistance systems and in autonomous vehicles, so the quality of the results and the processing times become critical, as detection failures or delays can have serious consequences. The obstacle detection system presented in this paper has been extensively tested during VIAC, the VisLab Intercontinental Autonomous Challenge [1], [2], which has offered a unique chance to face a number of different scenarios along the roads of two continents, in a variety of conditions; data collected during the expedition has also become a reference benchmark for further algorithm improvements. I. INTRODUCTION Fast methods to obtain accurate and dense depth maps are becoming increasingly common [3], [4], [5], and the problem of extracting useful information from such a big amount of data is of great interest. Even without taking into account algorithmic complexity, giving a definition of an obstacle is not a trivial task: an option is to determine a dominant ground surface [6] and consider as an obstacle anything sticking out of it; anyway, there are situations where this approach fails (e.g. very cluttered environments with no clearly visible road area, or lateral slopes). If no assumption can be done, as it was the case during the VIAC expedition, it is safer to consider the the ego-vehicle mechanics (e.g. height, width, maximum traversable slope) to identify areas that cannot be crossed. The downside of this approach is its computational weight, which limits the number of points that can be handled: nevertheless, this paper presents a parallel processing scheme which allows to run at 10 Hz on a commercial hardware platform. A. Hardware configuration Images used for stereo reconstruction are acquired at a resolution of 752 × 480 pixels by a pair of IEEE1394-A cameras equipped with 4 mm, 1/3 ” lenses; synchronization is guaranteed by a hardware trigger signal. The sensors are mounted right below the solar panel, as can be seen in Fig. 1-a. Processing is performed on a Mini-ITX board with an Intel ® Core™ 2 Quad Q9100 @ 2.26 GHz processor and 4 GB RAM located on the back of the van (Fig. 1-b). (a) (b) Fig. 1. Highlighted in red, hardware components of the stereo system: (a) the forward-looking stereo cameras, and (b) the processing unit in the back. B. The expedition During VIAC the vehicles crossed a number diverse en- vironments, ranging from country motorways in Hungary to busy downtowns in Russia, from the 2900 m Lanquan moun- tain pass to highway construction areas in China (Fig. 2); the weather also changed dramatically, from the hot summer of Ukraine (with an average temperature of 45 °C) to the cold September in Russia, the pouring rain of China, and snow on the mountain passes. Not all of what the vehicles had to face could be anticipated, but that was also the purpose of the test: to design an algorithm as robust as possible, evaluate its performance, and improve it afterwards using the data collected in the most critical scenarios. 2011 IEEE/RSJ International Conference on Intelligent Robots and Systems September 25-30, 2011. San Francisco, CA, USA 978-1-61284-455-8/11/$26.00 ©2011 IEEE 1599