CG Benefited Driver Facial Landmark Localization Across Large Rotation.

Liang Shi,Jiguang Yue,Yanchao Dong,Minjing Lin,Senbo Wang,Runjie Shen,Zhiming Chang
DOI: https://doi.org/10.1109/ivs.2018.8500537
2018-01-01
Abstract:Facial landmark localization is a crucial initial step Driver Inattention Monitoring. The aim of this paper is to localize driver facial landmarks across large rotation, say [−90°, +90°] in yaw rotation, to cope with real driving conditions. The paper proposes a flexible pipe-line for creating automatically labeled face image to supply wanted dataset. The benefits of CG (Computer Graphics) techniques such as 3D face modelling and morphing, photorealistic rendering and ground truth generation are utilized. To the best of our knowledge this is the first time to combine CG rendering and automatic ground truth labelling techniques with face landmark localization algorithms. The effectiveness of the CG rendered data is proved by cross validation with Multi-PIE dataset. Landmark localization across large rotation is obtained by a system simply integrating the off the-shelves algorithms and trained with the CG rendered data. The experiments of the implemented system on Multi-PIE and real persons show that it could localize facial landmarks across large rotation accurately and in real time.
What problem does this paper attempt to address?