中文摘要：

提出针对概率分布参数时变的噪声空间的双目视觉几何估计问题，由于Hartley三维重构几何估计算法是针对双目视觉受到同等噪声影响下所采用的．因此，对于受到不同噪声影响的情况而言，这个方法不能有效、准确地估计实时变化的测量点位置．通过构建动态贝叶斯网，利用先验和后验的知识进行预测和滤波，结合贝叶斯增量式学习方法并充分利用了其学习所获得的噪声样本空间概率模型变化演进的规律，这样可以较准确、平滑地估计出噪声对摄像头的影响，并以此来改进Hartley三维重构几何估计算法．

英文摘要：

The Geometric Estimate problem of computer vision was introduced according to variable noise sample space. The Hartley＇ s Geometric Estimate algorithm is adopted toward the same noise in the cameras. However,with the different noise,this algorithm is not suitable to estimate the real-time position of image points. The popular Dynamic Bayesian Network（DBN） was used as learner, predictor or filter. The unsupervised Bayesian learner,classifier and DBN were syncretized. It would acquire the roles that the probability or stochastic process model of noise evolved with time so as to the real-time noise of vision space can be estimated dynamic and accurate. Hartley＇algorithm is improved of it.