Energy-Transfer Features for Pedestrian Detection

In this paper, we propose an interesting and novel method for computing the image features that are useful for object detection. The method is interesting and novel in the terms of the feature vector dimensionality and object information capturing. In the proposed method, the areas of objects (that contain the important information useful for recognition) are described by the distribution of energy. The energy is transfered through the energy sources that are placed into the image and the distribution of energy is encoded into a vector of features. The vector is then used as an input for the SVM classifier. Using this approach, the objects of interest can be successfully described with a relatively small set of numbers if compared with the state-of-the-art descriptors that are based on the histograms of oriented gradients. We show the robustness of the features in the task of pedestrian detection.