Stable detection using an inner patch

Abstract

In this paper, we propose a novel method for stable detection of targets based on random patterns. The proposed method uses an inner patch for excluding ambiguity of random patterns. The proposed method consists of 2 steps. In the first step target candidates are classified at every position and target scale. In the second step the candidate which has the highest similarity to the target is selected. We use random Ferns for descriptors of the proposed method.