R. Kamada and T. Hayashi (Japan)
Object tracking, object extraction, interactive image cutout
This paper presents a silhouette-based object tracking method which has a mechanism of automatic tracking-error correction. The method starts with the process that a user draws seed lines in the objects and their background respectively at the ﬁrst frame of the video. The seed lines are
automatically tracked to the next frame with optical ﬂow estimation. In the line tracking process, the seed lines are automatically corrected when tracking error occurs. To the tracked seed lines in each frame, an image cutout method is used for partitioning the image into the object and back
ground. Object tracking can be achieved by repeating the seed line tracking from frame to frame and object extraction with image cutout. For the purpose of making clear the performance of the proposed method and the efﬁciency of the error correction in the line tracking process, this paper measures tracking accuracies of two tracking methods:
ones with/without the error correction. Experimental results have shown that the proposed tracking has high tracking accuracy because the error-correction works effectively.