自由立体显示中的视频图像在合成过程中存在速度要求高,图像处理数据量大等特点,在显示过程中,一旦用户偏离有效观看视区则无法看到正确的立体图像,到目前为止这些问题仍然存在,并阻碍着立体显示行业的发展。针对这种情况,提出了一种结合人眼跟踪算法的立体视频合成系统。通过判断人眼在屏幕前的位置,实时调整立体图像融合算法,使人眼始终能看到正确的立体图像对。由于该系统对跟踪的实时性和精确度要求较高,之前已有的算法很难同时在这两方面表现出色,因此提出了一种改进的人眼跟踪算法。该算法核心是基于ASM模型的人脸检测,并充分考虑了现场噪声和人体姿态频繁调整的特点,对ASM模型建立时空缓冲模型,处理结果直接映射到实际光栅空间分布模型中。根据当前ASM缓冲模型和光栅空间分布对应的映射点,对显示的图像进行相应的立体合成处理。这种自动跟踪人眼位置的立体视频显示方法跟踪速度快,位置计算精确,有效地扩大了立体视图区域。实验结果表明,该方法使观看者可以自由移动,在适当的范围内都可以看到清晰的立体影像,同时视频合成与播放的速度流畅,大幅提高了用户观看的舒适感。

3D video fusion for auto-stereoscopic displays has some difficulties, such as massive volumes of computation data, high-demand in the transition efficiency. Furthermore, when the viewers eyes are moving out of the effective stereo visual area, its hard to get the correct 3D video comfortably. These problems have not yet been well-solved so far and blocked the industrialization of auto-stereoscopic displays. Due to these facts ,we propose a method of multi-view stereo video fusion based on eye-tracking technique, which effectively enlarged the stereo visual area. Because of the high requirement of real-time and accuracy, the existing algorithms are hard to perform excellently in both respects, therefore an improved eye-tracking algorithm are proposed. This algorithm is based on ASM face detection with full consideration of the various noise and the frequent changes of the viewers postures. It establishes aspatial and temporal filter based on ASM, and maps the correct 3D images directly related onto the space distribution by the optical grating. According to these specific mapped points of ASM, it processes the fusion of the input 3D video images. Experimental results show that this method can allow the viewer move freely in front of the stereoscopic display. It could transmit the 3D video smoothly and effectively while complex computation and eye-tracking are carrying on in the background and it can improve the viewers comfort level significantly.