紀要論文

動的背景差分を用いる歩行者飛び出し検出システム

A System of Detecting Walkers Using Dynamic Background Subtraction

jpn

18810667

https://doi.org/10.15002/00010528

法政大学大学院情報科学研究科

9

95

100

2014-03

Version of Record

We proposed "Dynamic Background Subtraction" system for detecting walkers from mobile camera scenes, in order to prevent traffic accidents among automobiles and the walkers. This system separated the walkers from dynamic background of the scene. The system was based on the fact that front street scene of the mobile camera extended from one point: Infinite point. Analyzing the scene extensions, current scene was precisely predicted from the previous one. Stereo-camera provided us depth information for the scene prediction. The difference between the predicted scene and the real one removed background objects in Region of Interested, where the walkers run out into street in the scenes. In order to reduce the background noises made by vibration and/or curve movement of the automobile, the predicted scene was shifted and subtracted from the real one. Morphological operation indicated the walkers' residual. The proposed system was characterized by its simplicity in principle and high potentiality in easily realizing the system with low cost. The method was able to prevent the miss-extraction of background objects and un-extraction of walkers. In this paper, the principle and the procedures of the method were described, and the experimentation which detected walkers from mobile scenes by using the method showed its utility.

Article

AA12222297