Fast Facial Landmark Points Tracking Algorithm And Its Application In Raspberry Pi

Face landmark points tracking mainly through access to effective facial feature information from the face image, so we can take advantage of the information to do some personal identification or face expression recognition, at present, the main model method to obtain facial features information from face images are consist of active contour model, deformation template model and active statistical object model, active appearance model is a sort of active statistical model, which is an important feature extraction model algorithm. Due to the active appearance model has good versatility and flexibility, it is widely used in facial image processing, such as face recognition, face landmark points tracking and monitoring as well as facial expression analysis.In this paper, we do some research on face landmark points iterative alignment algorithm based on the Active Appearance Model, first the accuracy and speed of previous inverse compositional image alignment algorithm can not meet the requirement of the real-time face images processing, so in this paper, we come up with synchronization inverse compositional image alignment algorithm,combines the Active Appearance Model with synchronization inverse compositional image alignment algorithm, we make full use of the powerful modeling capabilities of the active appearance model and fast matching capabilities of fast synchronous inverse compositional image alignment algorithm, at the same time, we do some experiments to figure out the optimal number of training sample and the number of iterations in order to achieve a fast landmark points match, so as to realize the real-time facial recognition and landmark tracking, experimental results show that the algorithm can precise matching and tracking the landmark points in the face images from the video.Finally, base on the Raspberry Pi ARM platform system, we developed the face video tracking system on the Linux GUI development environment based on Active Appearance Model, the software platform embedded fast synchronous inverse compositional image alignment algorithm, and using independent set of face images samples collected, sample images are collected from the surrounding people, the system is consisted of landmark points marking module, active appearance model modeling module, active appearance model matching module, video tracking module and matching information storage module, final realize facial feature matching video tracking and face recognition tasks based on Active Appearance Model. This platform laied a solid foundation for further research on Active Appearance Model algorithm.