3D Face Modeling and Head Pose Measurement Using Stereovision

Abstract

To drive safely, it is effective to introduce the environmental recognition system such as the walker detector and the obstacle detector. It is, however, not preferable to show the driver all of information recognized by these systems, because it is difficult for the driver, especially for the elderly driver, to process too much information. Thus, it is necessary to eliminate the unnecessary information that displayed by the driving support system. It is considered that the oversight detection based on driver's gaze estimation will become powerful criterion for unnecessary information. The gaze position can be analyzed from both driver's head pose and gaze directions. This paper focuses on the head pose measurement technique in six degree of freedom. This technique is based on ICP registration between 3D face model made in advance and obtained shape data. ICP registration can achieve high accuracy alignment, but needs huge computational effort, in general. To solve this problem, this paper propose the introducing a reduction in volume of data based on pattern information in the image, coarse-to-fine search, and SIMD computing. Finally, it was confirmed that the proposed technique could measure in high accuracy by experiment.