Authors:
Katherine Batista
;
Rui Caseiro
and
Jorge Batista
Affiliation:
University of Coimbra, Portugal
Keyword(s):
Foreground segmentation, Shadow modelling and detection, Traffic surveillance.
Related
Ontology
Subjects/Areas/Topics:
Applications
;
Computer Vision, Visualization and Computer Graphics
;
Image and Video Analysis
;
Pattern Recognition
;
Segmentation and Grouping
;
Software Engineering
;
Video Analysis
Abstract:
This paper presents a method to automatically model and detect shadows on highway surveillance scenarios. This approach uses a cascade of two classifiers. The first stage of this method uses a weak classifier to ascertain the color information of possibly shadowed pixels which will be used by the second stage of this method (strong classifier). The weak classifier estimates the Color Normalized Cross-Correlation (CNCC) and the color information of the pixels identified as shadow, will be used to build or update multi-layered statistical shadow models of the RGB appearance of shadow. These models will then be used, by the strong classifier, to correctly distinguish shadow. To prevent misclassifications from corrupting the results of both classifiers, spatial dependencies are also taken into account. For this purpose, nonparametric kernel density estimators in a pyramidal decomposition (PKDE), as well as, Markov Random Fields (MRF) were independently employed. This technique is being u
sed in a real outdoor traffic surveillance system in order to minimize the effects of cast vehicle shadows as well as shadows induced by illumination changes. Several results are presented in this paper to prove its effectiveness and the advantages of applying spatial contextualization methods to the weak and strong classifiers.
(More)