Human Detection in Top-View Images Using Only Color Features

DOI Web Site

Bibliographic Information

Other Title
  • 上方から撮影された画像中における色特徴のみを用いた人検出
  • ジョウホウ カラ サツエイ サレタ ガゾウ チュウ ニ オケル イロ トクチョウ ノミ オ モチイタ ヒト ケンシュツ

Search this article

Abstract

Real-time vital sensing during exercise using sensor nodes attached to humans is a challenging problem because the density and the moving speed of the sensor nodes are very high. To solve this problem, the authors are trying to construct a novel routing scheme for multi-hop networking named “image-assited routing” that obtains locations of sensor nodes by image-based human detection. This paper shows that accurate detection required for the image-assisted routing can be achieved if top-view images are used for human detection. To evaluate detection accuracy in top-view images, a CG-based data set was construced using actual human motions during exercise. Experimental results using the constructed dataset showed that a detector trained with only color features selected by informed-filters achieved about 0.83% miss rate at 0.1 FPPI by exhausitive search based on sliding windows.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top