Real-time Eye Positioning System For Stereoscopic Display

Eye positioning is an important research issue in the field of machine vision, involved in digital image processing, pattern classification and machine learning. The human eye positioning plays a very important role on autostereoscopic display. In order to meet needs of autostereoscopic display, the system has the characteristics of real-time, high accuracy, high stability.We study common face detection and eye positioning algorithms and propose a customized algorithm for autostereoscopic display. The customized eye positioning algorithm has been already implemented on PC and a self-designed DSP platform. The system meets the needs of the correct rate, time consuming, face deflection and obstructions.First, the eye positioning system described in this paper introduces the background detection and bright pupil effect to filter the candidate regions. Second, Faces are found by Adaboost cascade classifier on the candidate regions. Third, the eye area is caught by an Adaboost+SVM approach. Last, pupil positioning algorithm is introduced to improve the detected accuracy. In addition, a motion-related filtering algorithm is presented to overcome the irregular jitter of the positioning result. The filter can determine if the current positioning result is valid in real time according to the correlation between positioning results and tracking results.We have designed a embedded platform based on TI TMS320DM6437 DSP to support the whole process, including receiving analog video frame, achiving eye positioning algorithm and sending the pupils data to the backlight module via RS232 interface. In order to complete involuted video-processing algorithms, the platform has the capability of fast communication, high process, high-speed large-capacity storage systems and rich interface. This paper introduces the optimization to get real-time requirement. Experiment results shows that the system is stable and accurate.