Real-time tracking of facial features (DEAL 684)

This lot is generally related to computer vision technology for the detection of facial features in images and tracking features in a series of images to accurately reconstruct deformable objects in images. Disclosed is a technique for detecting and tracking facial features in images containing large feature movements due to aspect changes from the motion of the subject or the motion of an imaging sensor. The system tracks detected features where there are large variations in shapes and appearances over time using clustered representations of non-linear shapes. The system identifies facial features in a target image by searching for the contours in the clustered shape subspaces using appearance information extracted at each landmark position and creating an initial shape in the target image. Further disclosed are techniques for superimposing the initial shape on the target image and iteratively deforming the shape until a final shape is produced. The technology may be implemented in feature tracking in images, object recognition in videos, face-tracking systems, multi-view face tracking systems, etc.