Abstract
Despite many years of research, obstacle recognition is still a difficult, but very important task. We present a multi-class approach, that extracts from images the Histogram of Oriented Gradients (HOG) based on aspect ratio of Region of Interest (ROI) and use them in a multiclass classification problem. For the learning phase we propose an original approach based on decision trees. Numerical experiments are performed on a benchmark dataset consisting of animal, pedestrian, car and sign (labeled) images captured in outdoor urban environments and indicate that the proposed model is able to improve the performance of the recognition process.
Citare
Mocan R., Dioșan L., Obstacle Recognition in Traffic by Adapting the HOG Descriptor and Learning in Layers, Studia Universitas Babes-Bolyai, Seria Informatica, 2015, LX(2):47-54
http://www.cs.ubbcluj.ro/~studia-i/contents/2015-2/04-MocanDiosan.pdf