IEEE Trans Image Process. 2010 Nov;19(11):3000-11. doi: 10.1109/TIP.2010.2050630. Epub 2010 May 18.
In the matching tasks of tracking and geometrical vision, there are usually priors available on the absolute and/or relative image locations of features of interest. In this paper, we use these priors dynamically to guide a feature by feature matching search that can achieve global matching with much fewer image processing operations and lower overall computational cost. First, the concept of dynamic sequential search (DSS) is presented. Then, the problem of determining an optimal search order for DSS is investigated, when the probabilistic distribution of the features can be described by a multivariate Gaussian model. Based on the general formulas for sequentially updating the predicted positions of the features as well as their innovation covariance, the theoretic lower bound for the sum of the areas of the features search-regions is derived, and the necessary and sufficient condition for the optimal search order to approach this lower bound is presented. After that, an algorithm for dynamically determining a suboptimal search order is presented, with a computational complexity of O(n3), which is two magnitudes lower than those of the state-of-the-art algorithms. The effectiveness of the proposed method is validated by both statistical simulation and real-world experiments with a monocular visual SLAM (simultaneous localization and mapping) system. The results verify that the performance of the proposed method is better than the state-of-the-art algorithms, with both fewer image processing operations and lower overall computational cost.
在跟踪和几何视觉的匹配任务中，通常可以利用特征的绝对和/或相对图像位置的先验知识。在本文中，我们动态地利用这些先验知识来指导逐个特征的匹配搜索，从而可以用更少的图像处理操作和更低的总体计算成本实现全局匹配。首先，提出了动态顺序搜索（DSS）的概念。然后，当特征的概率分布可以用多元高斯模型描述时，研究了确定 DSS 最佳搜索顺序的问题。基于特征预测位置以及其创新协方差的顺序更新的一般公式，推导了特征搜索区域总面积的理论下界，并给出了最佳搜索顺序接近该下界的必要和充分条件。之后，提出了一种具有计算复杂度为 O(n3)的动态确定次优搜索顺序的算法，比最先进的算法低两个数量级。通过单目视觉 SLAM（同时定位和制图）系统的统计模拟和实际实验验证了所提出方法的有效性。结果验证了所提出方法的性能优于最先进的算法，具有更少的图像处理操作和更低的总体计算成本。
