Machine Vision and Applications (2010) 21:555–576 DOI 10.1007/s00138-009-0185-z ORIGINAL PAPER An adaptive, real-time, traffic monitoring system Tomás Rodríguez · Narciso García Received: 15 October 2007 / Revised: 7 August 2008 / Accepted: 21 December 2008 / Published online: 27 January 2009 © Springer-Verlag 2009 Abstract In this paper we describe a computer vision-based traffic monitoring system able to detect individual vehicles in real-time. Our fully integrated system first obtains the main traffic variables: counting, speed and category; and then com- putes a complete set of statistical variables. The objective is to investigate some of the difficulties impeding existing traffic systems to achieve balanced accuracy in every condi- tion; i.e. day and night transitions, shadows, heavy vehicles, occlusions, slow traffic and congestions. The system we pres- ent is autonomous, works for long periods of time without human intervention and adapts automatically to the chang- ing environmental conditions. Several innovations, designed to deal with the above circumstances, are proposed in the paper: an integrated calibration and image rectification step, differentiated methods for day and night, an adaptive seg- mentation algorithm, a multistage shadow detection method and special considerations for heavy vehicle identification and treatment of slow traffic. A specific methodology has been developed to benchmark the accuracy of the different methods proposed. 1 Introduction The use of Intelligent Transport Systems (ITS) is progres- sively becoming more important for the efficient manage- ment of road traffic infrastructures. These systems comprise T. Rodríguez (B ) ETSI Informática, Universidad Nacional de Educación a Distancia, Madrid, Spain e-mail: tomasrod@yahoo.com N. García Grupo de Tratamiento de Imágenes, Universidad Politécnica de Madrid, Madrid, Spain a great number of technical aids and management strategies from different domains. In this paper we focus our attention on a set of advanced sensing technologies categorized under the term Intelligent Vehicle Highway Systems (IVHS). The aim is to describe an adaptive, real-time, computer vision- based, traffic monitoring system able to operate day and night in highways. Computer vision presents significant advantages over other more traditional vehicle measurement technologies (i.e. current loops). Computer vision systems are more flexible, less invasive, more precise, more robust, easier to maintain, produce richer information, do not affect the integrity of the road and offer as an added bonus, the possibility to transmit images for human supervision. Computer vision applied to traffic has been investigated since the late 1980s, but there is still intense research work going on. The bibliography dealing with this subject is huge. There are lots of good scientific publications [1–3] and a few very systematic reports on massive testing experiences [4–6]. It is also possible to find a certain number of commer- cial systems [7, 8]. Unfortunately, despite its undeniable interest, computer vision is not massively used in traffic monitoring applica- tions, since existing systems still suffer from poor reliability, high cost and unbalanced accuracy. The reason is computer vision systems are much affected by weather and illumination conditions. Their accuracy is seriously limited under chang- ing weather and it is not uncommon to find blocked systems when conditions are adverse. In addition, not every system is able to work unattended 24 h; i.e. many systems cannot operate at night or in the periods between day and night. On the other hand, most systems are unable to cope with slow traffic or congestions and show important difficulties to correctly detect heavy vehicles. Occlusions and shadows are also important problems, causing lots of errors. 123