Cascaded Continuous Regression for Real-Time Incremental Face Tracking

This paper introduces a novel real-time algorithm for facial landmark tracking. Compared to detection, tracking has both additional challenges and opportunities. Arguably the most important aspect in this domain is updating a tracker’s models as tracking

  • PDF / 1,819,946 Bytes
  • 17 Pages / 439.37 x 666.142 pts Page_size
  • 112 Downloads / 242 Views

DOWNLOAD

REPORT


Abstract. This paper introduces a novel real-time algorithm for facial landmark tracking. Compared to detection, tracking has both additional challenges and opportunities. Arguably the most important aspect in this domain is updating a tracker’s models as tracking progresses, also known as incremental (face) tracking. While this should result in more accurate localisation, how to do this online and in real time without causing a tracker to drift is still an important open research question. We address this question in the cascaded regression framework, the state-of-the-art approach for facial landmark localisation. Because incremental learning for cascaded regression is costly, we propose a much more efficient yet equally accurate alternative using continuous regression. More specifically, we first propose cascaded continuous regression (CCR) and show its accuracy is equivalent to the Supervised Descent Method. We then derive the incremental learning updates for CCR (iCCR) and show that it is an order of magnitude faster than standard incremental learning for cascaded regression, bringing the time required for the update from seconds down to a fraction of a second, thus enabling real-time tracking. Finally, we evaluate iCCR and show the importance of incremental learning in achieving state-of-the-art performance. Code for our iCCR is available from http://www.cs.nott.ac.uk/∼psxes1.

1

Introduction

The detection of a sparse set of facial landmarks in still images has been a widely-studied problem within the computer vision community. Interestingly, many face analysis methods either systematically rely on video sequences (e.g., facial expression recognition [1]) or can benefit from them (e.g., face recognition [2]). It is thus surprising that facial landmark tracking has received much less attention in comparison. Our focus in this paper is on one of the most important problems in model-specific tracking, namely that of updating the tracker using previously tracked frames, also known as incremental (face) tracking. The standard approach to face tracking is to use a facial landmark detection algorithm initialised on the landmarks detected at the previous frame. This Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46484-8 39) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016  B. Leibe et al. (Eds.): ECCV 2016, Part VIII, LNCS 9912, pp. 645–661, 2016. DOI: 10.1007/978-3-319-46484-8 39

646

E. S´ anchez-Lozano et al.

Fig. 1. Overview of our incremental cascaded continuous regression algorithm (iCCR). (a) shows how continuous regression uses all data in a point’s neighbourhood, whereas sampled regression uses a finite subset. (b) shows how the originally model RT learned offline is updated with each new frame.

exploits the fact that the face shape varies smoothly in videos of sufficiently high framerates: If the previous landmarks were detected with acceptable accuracy, then the initial shape will be close enough for