A study on pedestrian detection using a deep convolutional neural network

Pedestrian detection is presently a topic of interest in computer vision due to its applications as an aid in car driving and in surveillance. The good results obtained using Convolutional Networks for vision tasks make them an attractive tool to improve the capabilities of pedestrian detection systems. In this work we study the use of a Convolutional Network as a refinement for classification of candidate regions previously detected using Haar features embedded in an AdaBoost scheme. The data used for training and testing come from the INRIA pedestrian database. The influence of design parameters, such as, the number of stages of the cascade in the detection stage and the scale factor in the pyramid of the multi-scale method, have been studied.