Self-supervised learning method for unstructured road detection using Fuzzy Support Vector Machines

• Published in: 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems pp. 1183 - 1189
• Main Authors: Shengyan Zhou, Iagnemma, K
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2010
• Subjects: Classification algorithms; Feature extraction; Pixel; Roads; Support vector machines; Training; Training data
• ISBN: 9781424466740; 1424466741
• ISSN: 2153-0858
• DOI: 10.1109/IROS.2010.5650300

Road detection is a crucial problem in the application of autonomous vehicle and on-road mobile robot. Most of the recent methods only achieve reliable results in some particular well-arranged environments. In this paper, we describe a road detection algorithm for front-view monocular camera using road probabilistic distribution model (RPDM) and online learning method. The primary contribution of this paper is that the combination of dynamical RPDM and Fuzzy Support Vector Machines (FSVMs) makes the algorithm being capable of self-supervised learning and optimized learning from the inheritance of previous result. The secondary contribution of this paper is that the proposed algorithm uses road geometrical assumption to extract assumption based misclassified points and retrains itself online which makes it easier to find potential misclassified points. Those points take an important role in online retraining the classifier which makes the algorithm adaptive to environment changing.