Observing on-road vehicle behavior: Issues, approaches, and perspectives

In this work, we review recent works comprising an emerging field of intelligent transportation: behavior analysis of vehicles. The ITS community has approached this topic both from vehicle-based and infrastructure-based sensing. In both cases, motion is the key indicator required for behavioral characterization, with accurate long-term prediction being the ultimate goal. However, the popular methods for behavior characterization differ between the sensing methodologies. Vehicle-based sensing tends to focus on spatio-temporal measurements coupled with various features for accurate estimation of object state. In contrast, infrastructure-sensing tends to avoid attempting high resolution estimation of vehicle state and prefers to utilize patterns learned in aggregate for constrained estimation. This review focuses on vision-based sensing and provides highlights of state-of-the art methods used in surveillance, and on-road vision modalities. We provide discussion and comment on future directions in the field.