The Community for Technology Leaders
Green Image
This paper proposes a novel unsupervised algorithm learning discriminative features in the context ofmatching road vehicles between two non-overlapping cameras. The matching problem is formulated as asame-different classification problem, which aims to compute the probability of vehicle images from twodistinct cameras being from the same vehicle or different vehicle(s). We employ a novel measurementvector that consists of three independent edge-based measures and their associated robust measurescomputed from a pair of aligned vehicle edge maps. The weight of each measure is determined byan unsupervised learning algorithm that optimally separates the same-different classes in the combinedmeasurement space. This is achieved with a weak classification algorithm that automatically collectsrepresentative samples from same-different classes, followed by a more discriminative classifier basedon Fisher' s Linear Discriminants and Gibbs Sampling. The robustness of the match measures and the useof unsupervised discriminant analysis in the classification ensures that the proposed method performsconsistently in the presence of missing/false features, temporally and spatially changing illuminationconditions, and systematic misalignment caused by different camera configurations. Extensive experimentsbased on real data of over 200 vehicles at different times of day demonstrate promising results.
Object recognition, unsupervised learning, Gibbs Sampling, Fisher' s Linear Discriminants, edge feature, vehicle matching, object reacquisition, non-overlapping camers.

Y. Shan, H. S. Sawhney and R. Kumar, "Unsupervised Learning of Discriminative Edge Measures for Vehicle Matching between Nonoverlapping Cameras," in IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 30, no. , pp. 700-711, 2007.
82 ms
(Ver 3.3 (11022016))