【作者】 邱阳； 白立芬； 李庆祥； 徐毓娴；

【Author】 Qiu Yang Bai Lifen Li Qingxiang Xu Yuxian (Department of Precision Instrument and Mechanology,Tsinghua University , Beijing 100084)

【机构】 清华大学精密仪器与机械学系；

【摘要】 从二维图像恢复三维信息的立体视觉研究已经出现了许多基于边缘和基于区域的方法,但对硬件的要求制约着它们的实用性。本文提出了一种对边缘和区域进行联合匹配的算法,采用边缘信息进行粗糙匹配以提高处理速度,得到一系列匹配的可能位置作为精确匹配的初始值域;采用区域图像进行精确匹配,以得到更多的细节信息,而且其平均效应会减小高频噪声对运算的影响。建立的实验系统采用奔腾133CPU、16兆内存、商用标准CCD和采集卡,软件用VC6.0编制。模拟实验表明,系统运行正常,适应性良好。更多还原

【Abstract】 Based on the relationship between boundary and area,one idea to reconstruct 3D object from 2D perspective image is brought out in this paper that the combining of two characters will take more advantages in reducing the requirements for hardware. The correlation is programmed in two steps, pre-correlation and accurate correlation. The quickly pre-correlation of images is based on boundary, which provides accurate correlation an o-riginal value range. The accurate correlation based on area,with more details and fewer sensitivities to high frequency noise, leads to final solution. We carry it into effect under a limited hardware: Pentium 133 CPU,16M RAM,CCD and A/D adapter in commercial quality etc. Software is coded in VC 6. 0. Simulative experiments show that system performs normally in many cases.更多还原