Vehicle Classification with Confidence by Classified Vector Quantization

Abstract

Automated vehicle classification based on static images is highly practical and directly applicable for various operations such as traffic related investigations. An integrated vehicle detection and classification system is proposed in this paper. A multi-resolution vehicle detection scheme is introduced as an improvement over the cascade boosted classifiers proposed recently by Negri et al. 2008 in the literature. Building on solutions from previous works from Negri et al, the implementation of a new decision strategy renders current detection method to be robust to environmental changes. The vehicle classification is based on the Classified Vector Quantization (CVQ) proposed earlier by Zhang et al. 2009. The justification of choosing CVQ is its advantages in providing classification confidence by incorporating rejection option. The significance of rejection in enhancing the system?s reliability is emphasized and evaluated. A database composed of more than 2800 images of four types of vehicles (cars, vans, light trucks and buses) was created using police surveillance cameras. The proposed scheme offers a performance accuracy of over 95% with a rejection rate of 8%, and reliability over 98% with a rejection rate of 20%. This exhibits promising potentials for implementations into real-world applications.