F-040 Vehicle-borne Pedestrian Detection by Fusing Horizontal Laser Data and Video Images
スポンサーリンク
概要
- 論文の詳細を見る
This paper describes a system that uses horizontal laser data to improve image-based pedestrian detection. Our image-based pedestrian detection is an extension of Depoortere et al.'s work. We build a 4-layer cascade of classifiers using both Haar wavelet features and Histograms of Oriented Gradients (HOG) features. Our main contribution is to use horizontal laser data to estimate probable positions of pedestrians on the ground plane, which limits pedestrians' searching range in an image to a small area and certain scales. We attempt to apply a laser-based SLAM system on our intelligent vehicle to further reduce the pedestrians' searching area to regions of moving objects and newly observed objects. Our system has significantly lower false positive rate and computation cost, compared to image-based pedestrian detection. To the best of our knowledge, this is the first trial to combine the laser-based moving object detection with image-based pedestrian detection.
- FIT(電子情報通信学会・情報処理学会)推進委員会の論文
- 2010-08-20
著者
-
Zha Hongbin
The State Key Lab Of Machine Perception Peking University
-
Shibasaki Ryosuke
The Center For Spatial Information Science University Of Tokyo
-
Zhang Quanshi
The Center for Spatial Information Science, University of Tokyo
-
Zhao Huijing
The State Key Lab of Machine Perception, Peking University
-
Shao Xiaowei
The Center for Spatial Information Science, University of Tokyo
-
Shao Xiaowei
The Center For Spatial Information Science University Of Tokyo
-
Zhang Quanshi
The Center For Spatial Information Science University Of Tokyo
-
Zhao Huijing
The State Key Lab Of Machine Perception Peking University