Small face detection using deep learning on surveillance videos

Abstract

Face detection is one of the essential tasks widely studied in the field of Computer Vision. Several authors have developed different techniques to improve the face detection in images, but these are limited in their application on videos and more if they present low resolution. In this study, we propose a new model for face detection in low-resolution videos based on the morphology of the upper body of people, and the use of Deep Learning (CNN). Our results show an average of 39% accuracy over the Caviar dataset and 32% in the UCSP dataset. Compared with other techniques, our results are greater due they only reach 1% of accuracy.