에지 방향성 히스토그램을 이용한 실시간 물체 인식 시스템

Abstract

In this study of object recognition, The edge of all pixel can be get after elimination of noises on the screen by the edge orientation histogram. And with the threshold value, the decisions are made(follows by decision of edge for distinguishing) whether those pixels are edge or not. After generating the edge orientation histogram on the decided pixels, object recognition should be done in comparison between the edge orientation histogram, driven by mapping on the 36 saving area of all those pixels produced, and model screen. It's very sensitive to the variation of lights and the size of the subjects because the physical recognition of edge orientation histogram generally depends on the distribution of those pixels. Therefore, it requires a lot of practices, and the video images need to be processed with normalization to make their size even. The edge orientation histogram use the value of the pixel image so the process is very sensitive to the noises. In this paper, noises has been detected and removed to retain a certain recognition level.

The method, using the edge of the histogram suggested by this these, performs very fast, and produces reliable recognition rate in object recognition. However, because it is using the value of the pixel, in a complex background conditions, like with objects with color or objects in the video stacking, there is a lot difficulties in recognition. There is a plan to solve these problems and to develop more stable object recognition algorithms in the future.