Face Recognition by 3D Registration for the Visually Impaired Using a RGB-D Sensor

To help visually impaired people recognize people in their daily life, a 3D face feature registration approach is proposed with a RGB-D sensor. Compared to 2D face recognition methods, 3D data based approaches are more robust to the influence of face orientations and illumination changes. Different from most 3D data based methods, we employ a one-step ICP registration approach that is much less time consuming. The error tolerance of the 3D registration approach is analyzed with various error levels in 3D measurements. The method is tested with a Kinect sensor, by analyzing both the angular and distance errors to recognition performance. A number of other potential benefits in using 3D face data are also discussed, such as RGB image rectification, multiple-view face integration, and facial expression modeling, all useful for social interactions of visually impaired people with others.