Appearance-based Classification of Moving Cars by Eigendimension Matching

An appearance-based eigenspace method for classifying moving cars is introduced in this paper. A car can be classified by observing any of its partial view, using this method. We present a new algorithm called eigendimension matching instead of calculating Euclidean distance between eigensubspaces for the identification and classification of moving cars. The proposed method is based on the establishment of eigenspaces of a set of known car images, and comparing their eigendimensions to classify the unknown car or non-car images. Experimental results from the real-world data show the effectiveness of the proposed method