一种弱纹理图像特征跟踪的鲁棒方法

目的 提出一种用于弱纹理图像的特征提取和跟踪的鲁棒方法。方法 选取包含若干表面片的结构化特征，并假设结构化特征中的表面片之间的关系在图像运动时保持不变，使用基于梯度的方法和基于相关性的方法在另一幅图像中求对应的结构化特征，利用结构化特征不变性判据，评价对应结构化特征的有效性。结果与结论 该算法能有效地跟踪一般表面片特征，非常适合于在室外获取的一般图像序列和弱纹理图像序列运动分析。

A Robust Algorithm of Feature Point Tracking for Image Sequence

Aim To present a robust method for tracking feature and evaluating the tracking quality of an image sequence. Methods First, a structural feature was selected to contain several patches. The topology among the patches in the structural feature was assumed to be invariant. Then, the gradient based and correlation based algorithm is used to find out the correspondences of the patches between previous frame and current frame. Thirdly, the evaluation of correspondences under the constraint of invariant structure feature during motion was achieved. Results and Conclusion The approach was applied to several testing images and showed better tracking results than non structural feature algorithm.