A probabilistic distribution approach for the classification of urban roads in complex environments

Giovani Bernardes Vitor (1, 2) , Alessandro C. Victorino (1) , Janito V. Ferreira (2)

1 Heudiasyc - Heuristique et Diagnostic des Systèmes Complexes [Compiègne]
2 UNICAMP - Universidade Estadual de Campinas = University of Campinas

Navigation in urban environments has been receiving considerable attention over the past few years, especially for self-driving cars. Road detection for Autonomous Systems, and also for ADAS (Advanced Driving Assistance Systems) remains a major challenging in inner-city scenarios motivated by the high complexity in scene layout with unmarked or weakly marked roads and poor lightning conditions. This paper introduces a novel method that creates a classifier based on a set of probability distribution. The classifier, created using a Joint Boosting algorithm, aims at detecting semantic information in roads. This approach is composed of a set of parallel processes to calculate the superpixel using the Watershed Transform method and the construction of feature maps based on Textons and Disptons. As a result, a set of probability distribution is generated. It will be used as an input to model the week classifier by our Joint Boosting algorithm. The experimental results using the Urban-Kitty benchmark are comparable to the state-of-the-art approaches and can largely improve the effectiveness of the detection in several conditions.

Mots clés: Computer Vision; Texton Map; Dispton Map; Watershed Transform; Joint Boosting; Road Detection

Domaines: Robotique [cs.RO]

Giovani Bernardes Vitor, Alessandro C. Victorino, Janito V. Ferreira. A probabilistic distribution approach for the classification of urban roads in complex environments. IEEE Workshop on International Conference on Robotics and Automation, May 2014, Hong Kong, Hong Kong SAR China. ⟨hal-01089086⟩