ドライブレコーダーからの歩行者認識技術の研究

Abstract

The total number of accident itself and injured people are still numerous, although the number of casualty has been decreasing. Many studies and researches have been reported to identify/recognize pedestrians by using onboard video camera to prevent accidents. However current drive assist systems which identify pedestrians automatically and make warning are expensive in general. They are limited to install on luxury cars as optional equipment. This paper proposes to recognize pedestrians from the movie taken by an inexpensive drive recorder. It is difficult to accurately recognize pedestrians from drive recorder images since the resolution and quality of them is low. We apply support vector machine identifier to reduce false positives and provide accurate recognition. This identifier uses HOG(Histograms of Oriented Gradients) features which is effective to human recognition.