【摘要】 给出了一种面向增强现实(augmented reality,AR)的基于自然纹理特征的实时跟踪算法,算法分为目标检测与跟踪两个过程。以真实场景中的目标物体的表面纹理图像作为模板,用基于朴素贝叶斯分类的宽基线匹配方法进行目标检测与方位参数估计;将分层L-K光流算法与鲁棒的IC算法结合,提出一种基于角点与纹理的混合跟踪算法,并用于其跟踪过程。实验结果表明,所提算法具有较好的实时性、准确性与鲁棒性,并解决了宽基线匹配算法在AR应用中出现的抖动现象。更多还原

【Abstract】 A natural texture-based real-time tracking approach to augmented reality is presented,which includes detecting procedure and tracking procedure.Giving a texture image template of the object in real scene,the object detecting and pose estimate are achieved by using a wide baseline matching approach based on a naive Bayesian classifer.Combining a hierarchical L-K optical flow algorithm and a robust inverse compositional algorithm,a tracking algorithm based on corner amd texture features is proposed,which is applied in the tracking procedure above.Experiment results demonstrate that the presented approach is faster,more accurate and robust,and can avoid the jitter of wide baseline matching in AR application.更多还原