摘要：

根据增强现实在视频对象跟踪中的应用需求，提出一种综合利用尺度不变特征变换(SIFT)算子、K聚类算法和轮廓检测的视频对象跟踪算法。该算法利用简易SIFT获得输入图像的特征点，通过K聚类算法获得可能的对象聚类，并采用改进的轮廓处理方法得到对象边界，移除孤立点，确定对象特征点，在对象特征点中获取增强现实应用中需要的注册点。在关键帧匹配中，只要使用对象特征点进行对象匹配。实验结果表明，该算法具有运行速度快、匹配正确率高的特点，能满足增强现实视频应用的注册需求。

关键词: 增强现实, 视频对象跟踪, 尺度不变特征变换算子, K-means算法, 轮廓检测

Abstract:

According to the application requirement of Augmented Reality(AR) in video object tracking, this paper proposes a video object tracking algorithm based on Scale-Invariant Feature Transform(SIFT) operator, K-means clustering algorithm and contour detection. The reduced SIFT is applied to get the feature points from the input image. The K-means clustering algorithm is applied to cluster the object feature points approximatively. The improved contour process is applied to get outlines from the clustered object feature points, removes isolation points and determines the object feature points. The registered point is got from the object feature points set. In the key frame, it only needs to use the object feature points to match the object. Experimental results show that the algorithm is fast and accurate. It can meet the need of AR registering.

Key words: augmented reality, video object tracking, Scale-Invariant Feature Transform(SIFT) operator, K-means algorithm, contour detection

中图分类号: 

• TN911.73