Feature Extraction for Pedestrian Classification Under the Presence of Occlusions Laurens van der Maaten * Guido de Croon * MICC, Maastricht University, P.O. Box 616, 6200 MD Maastricht, The Netherlands {l.vandermaaten,g.decroon}@micc.unimaas.nl Abstract The identification of pedestrians is an important problem in a wide range of computer vision applications. Previous work on pedestrian identification mainly focuses on fea- ture extraction approaches. In this study, we investigate four new features for pedes- trian identification that aim to overcome the limitations of features that are employed in previous studies. In particular, we focus on features that are to some extent robust to the presence of occlusions in the pedestrian images. The four new feature extraction approaches that are investigated are based on: (1) nonlinear dimensionality reduction, (2) texton frequency histograms, (3) Latent Dirichlet Allocation, and (4) local receptive fields networks. We test the performance of our feature extraction approaches on the NiSIS competition dataset. From the obtained results, we conclude that Support Vector Machines trained on PCA features perform best, most likely due to the exploitation of biases in the NiSIS dataset. 1 Introduction The identification of pedestrians (or in general, persons) is an important problem in computer vision applications such as surveillance, robotics, and control systems. The overall appearance of an observer pedestrian is subject to a variety of transformations, such as viewing angle, lightning conditions, physical properties of the pedestrian, and occlusions. Humans are known to perform very well on the identification of pedes- trians under these transformations and distortions. In order to foster the development of robust techniques for pedestrian identification, NiSIS organized a competition that aims at the development of such techniques. In particular, the development of biologi- cally plausible techniques is encouraged. The number of previous studies on pedestrian classification are limited [9, 16, 18]. An approach to pedestrian classification using two high-definition cameras is presented by Scotti et al. [18]. The approach by Scotti et al. identifies pedestrians by means of their specific geometric properties. In [9], Gavrila and Munder present an approach to pedestrian identification that detects pedestrians by combining multiple cues such 1