An Eye-Tracking System Implemented On DSP Platform For Glasses-Free3D

Because of the film Avatar had obtained a great success,3D display technology has attacked people’s eyes deeply as fast as possible. It sounds that3D technology will be a new hot field for researchers to explore since the full HD display technology has shocked the high taste customers. So far there are two types of3D:Glasses-Free3D and Glasses3D. The former one can benefit audiences without a pair of glass compared with the latter type. For Glasses-Free3D, the most important technology is locating the eyes of audience. This thesis aims at researching an embedded eye-tracking system for Glasses-Free3D.A coarse-to-fine approach to eye-tracking is proposed in the thesis, which can directly divide the study into two parts. A robust face tracking is the first step then the eye-detecting on the face area that tracked and abstracted can be applied to reach the goal. In first part, this thesis proposed a coupled CamShift approach based on MeanShift combined with Kalman filter to realize face tracking. For eye-detecting, before utilizing support vector machine (SVM) to check out the real eye pair this thesis proposed a multi-neighborhood blocks with weight approach to form an eye candidates set to be verified. The proposed method can abstract less eye candidates as fast, accuracy and efficiency as possible which greatly reduced the computing-load.This thesis finally selected the DM6467from Texas Instruments (TI) company as the processing kernel. Firstly, the entire eye-tracking algorithm will be verified with OpenCV and Matlab. Then this algorithm will be realized in C code to transplant to the DSP based embedded system. Taking the real-time into consideration, the C code will be optimized according to the character of the DSP platform.