MASTER Face detection with deformable part models on the NVIDIA Tegra K1 Face Detection with Deformable Part Models on the NVIDIA Tegra K1

Disclaimer This document contains a student thesis (bachelor's or master's), as authored by a student at Eindhoven University of Technology. Student theses are made available in the TU/e repository upon obtaining the required degree. The grade received is not published on the document as presented in the repository. The required complexity or quality of research of student theses may vary by program, and the required minimum study period may vary in duration. General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain Abstract-In recent years, face detection has become a popular application in embedded devices. However, since computational and energy efficiency are key aspects when targeting embedded devices, most face detection algorithms sacrifice accuracy for efficiency. The Deformable Part Models (DPM) algorithm is known to greatly outperform integral-channel based detectors in terms of robustness, although at an increased computational cost, making DPM impractical for use on embedded devices. This paper proposes a method to speed up a DPM face detection algorithm when targeting the NVIDIA Tegra K1 SoC. Several strategies are proposed to offload the most computationally intensive algorithmic components to the GPU. In order to test the functional correctness of our implementation, we trained a three-component DPM face model and tested it on three challenging video sequences containing multiple persons. When comparing our results to a single-thread CPU implementation, we achieve a 1.5x speed-up without any loss in accuracy. I. INTRODUCTION Face detection is a mature topic in the domain of computer vision. It is used in many consumer products such as smartphones, digital cameras and personal computers, but it also forms a basis for more sophisticated computer vision algorithms, such as facial recognition systems. However, building a robust face detection system still remains a challenge because of the many visual variations that can occur in a video, such as illumination changes, pose and scale variations, or object rotation. Many of the current face detection systems have difficulties coping with these challenges. The Deformable Part Models (DPM) algorithm, first proposed by P. Felzenszwalb et al. While several solutions have been proposed to overcome the computational bottlenecks of DPM Contribution: This paper proposes a computational efficient GPU-based implementation of the DPM face detector on the NVIDIA Tegra K1. We provide an accurate analysis of th