部位スコアマップ統合による隠れにロバストな人体検出

Bibliographic Information

Other Title
  • ブイ スコアマップ トウゴウ ニ ヨル カクレ ニ ロバスト ナ ジンタイ ケンシュツ
  • Parts Score Map Integration for Partially Occluded Humans Detection

Search this article

Abstract

We propose a human body detection framework based on parts scope map integration, which is robust to partially occluded human image. Generally part based body detection methods utilize part detectors information such as presence or absence within certain region or their positions, by integrating them using e.g. probability distributions model. However partial body regions itself tend to be not having distinctive feature like whole body and it likely to occur false positives in case of e.g. complex background region. Another problem is deterioration of performance in case of major parts occlusion. In this paper, we propose (1) score map integration method, (2) multilayered integration method to address partial occlusion. (1) enables to distinguish part detection result of human body from them of non human body more clearly. (2) increases detection performance for partial occluded images. The evaluation result shows better performance especially in case of partial occlusion images in comparison to the traditional method.

IEICE Technical Report;PRMU2011-248,IEICE Technical Report;HIP2011−76

Journal

References(48)*help

See more

Details

Report a problem

Back to top