Abstract
This paper proposes high-speed image matching method using a template autonomously optimized by learning of detection history. While many template matching methods have been proposed as one of object detection method, there are some methods that optimizes a template image by machine learning. However, those methods need a long time for preprocessing to optimize the template image. Moreover, acquisition of large amount of data for machine learning is not easy. To solve this problem, we propose a method that optimizes a template image by executing detection and learning history simultaneously. Through experiments, it is shown that the template image can be optimized even if large amount of data is not prepared in advance.
