Cascaded classifiers are frequently used for object detection tasks. The base classifiers in a cascaded classifier can be of any type but the types of features used by these classifiers have a strong impact on the performance of the cascaded classifier. In this paper we present a method for the extraction of multiple features inside a rectangular region relying on integral images. The rectangular regions are chosen in a classifier training process and are sensitive to the geometrical properties of an object while the features extracted inside the regions are sensitive to structural characteristics. In combination with a cascaded classifier this results in a large variety of features sensitive to the geometrical as well as structural properties. Our experimental evaluation shows that a combination of both tensor and intensity based features with a specific normalization method yields the best performance of the examined approaches.
Resource optimized cascaded perceptron classifiers using structure tensor features for us speed limit detection
Möglichkeitsoptimierter, kaskadierter Perzeptron-Klassifizierer mit Hilfe von Tensorstruktur-Merkmalen für den Einsatz der US-Geschwindigkeitsbegrenzung
2010
6 Seiten, 4 Bilder, 3 Tabellen, 20 Quellen
(nicht paginiert)
Aufsatz (Konferenz)
Datenträger
Englisch
Optimization design of cascaded classifiers
IEEE | 2005
|The Perceptron-Based Hierarchical Structure of Classifiers Constructed by the Adaptive Method
British Library Online Contents | 1997
|On-board Pedestrian Detection by the Motion and the Cascaded Classifiers
Springer Verlag | 2011
|Multi-class Object Detection in Vision Systems Using a Hierarchy of Cascaded Classifiers
British Library Conference Proceedings | 2006
|