View independent vehicle/person classification

Abstract

In this paper, we present an object classification system for digital video surveillance which can be used for an arbitrary camera viewpoint. The system was designed to distinguish humans from vehicles for an arbitrary scene. The system employs a two phase approach. In the first phase, human/vehicle recognition is performed using classical feature-based classification. This phase is used to initialize view-normalization parameters. The parameters allow the second phase, to perform improved classification based on normalized features. The normalization also enables absolute identification of size and speed which can be used in various ways including identifying vehicles of a certain size and searching for objects traveling at specific speeds across different locations in the image and across different viewpoints/cameras. Copyright 2004 ACM.