Paper Details

TRB 90th Annual Meeting (January 23-27, 2011)

Event Number:	390
Event Title:	Intelligent Transportation Systems Development and Applications
Event Date:	Jan 24 2011 7:30PM- 9:30PM
Event Location:	Marriott, Salon 2
Event Description:
Event Agenda:

Method for Preceding Vehicle Type Classification Based on Sparse Representation (11-1724)

Preceding vehicle recognition is an important enabling technology for developing a driving assistance system and an autonomous vehicle system. A new vehicle type classifier based on sparse representation model, is developed in this paper for classification of preceding vehicles. Unlike general supervised learning methods, where a training procedure is used to create a classification model for testing, the sparse representation approach does not contain separate training and testing stages. Instead, classification is achieved directly out of the testing sample¡¯s sparse representation in terms of training samples. Another unique feature of the new method is no model selection needed. The implementation of sparse representation classification for vehicle type (SRCVT) includes the following four steps: data preparation, principal component analysis (PCA) transformation, realization of SRCVT and classification output. Numerical experiments are designed to quantitatively verify the performance of SRCVT using the vehicle region of interest (ROI) data gotten by the preceding vehicle detection process. Experiment results show that the proposed SRCVT approach can outperform the best performance achieved by SVMs and has no need of model selection.

Authors
     Chong, Yanwen , Wuhan University, China
     Chen, Wu , Hong Kong Polytechnic University
     Li, Zhilin , Hong Kong Polytechnic University
     Lam, William H. K. , Hong Kong Polytechnic University