Embedded Landmark implementation for Deep Learning pre-processing

Abstract

Due to the evolution of information technology, it is becoming increasingly easy to use new platforms in order to set up efficient systems that are well adapted to the expected needs. As part of improving security and facilitating the detection of potentially dangerous persons, an intelligent application for on-board facial recognition is being developed. It is within this framework that we propose this paper. The objective of the proposed work is twofold. On the one hand, we propose to develop a module for the detection of relevant facial characteristics, which is the first step of an intelligent video surveillance application. Based on the detection of points of interest of the Landmark algorithm, a software optimization of the work is proposed. On the other hand, this application will be decomposed in order to be embedded on a multiprocessor architecture. In order to validate the multiprocessor-based approach, a comparison with other existing powerful processor architectures will allow to validate the best approach. This work will be the input for an intelligent embedded face detection application based on Machine Learning.