New Deep Convolutional Neural Network Architecture for Pedestrian Detection

JACOBO JULIO; I. OROZCO; BUEMI, M. E.

Madrid

Conferencia; International Conference on Pattern Recognition Systems (ICPRS-17); 2017

Universidad Carlos III

Pedestrian detection is currently a topic of interest in computervision due to its applications such as driver assistance systemsand surveillance in public spaces, among others. The goodresults obtained using deep convolutional networks in visiontasks make them an attractive tool to improve the capacities ofpedestrian detection systems. In this work we propose a deepconvolutional network architecture to classify as pedestrian ornon-pedestrian the candidate regions previously generated us-ing a simple pyramidal sliding window approach. A distin-tive characteristic of the CNN in this system is that it sepa-rates pedestrian from non-pedestrian images without the aid ofa pre-classification stage, and without the need of special tun-ing steps or initials conditions, making it more straightforwardthan other CNN-based solutions.The data used for training andtesting come from the Caltech-USA Pedestrian dataset [8, 9].We have evaluated the classification results on the proposedarchitecture and have obtained an average of ∼ 98% successfor the validation set. We have also carried out another evalu-ation of our system using the Benchmark proposed by Dolláret.al. [9], and obtained results that are competitive with thosementioned in the bibliography.