Background suppressed shapelet features used for vehicle detection

In this paper, we address the problem of detecting vehicles in road surveillance videos. An algorithm for learning background suppressed shapelet features, a set of middle-level features, is proposed. This new algorithm is based on motion detection [1-10] and learning shapelet features [18]. First, moving objects are detected; then, background suppressed shapelet features are extracted to train a classifier which classify those objects into two classes: vehicles and non-vehicles. The difference between background suppressed shapelet features and original shapelet features is that the former one utilizes some background suppression techniques to suppress the gradients in the background areas to eliminate the influence of those gradients while the latter one does not. Experimental results show that our algorithm outperforms the original shapelet algorithm.