Citation

Kim HK, Takahashi T, Kamijo S. Trans. Soc. Automot. Eng. Jpn. 2013; 44(2): 573-578.

Copyright

(Copyright © 2013, Society of Automotive Engineers of Japan)

DOI

10.11351/jsaeronbun.44.573

PMID

unavailable

Abstract

This paper presents a system that detects pedestrians and categorizes their "class". Foreground objects are detected as Regions of Interest (ROI) by a linear approximation on background motion vectors and tracked by the S-T MRF model. Then, a cascade of heterogeneous classifiers scans inside the ROI in which area a pedestrian exists. Finally, our system categorizes each ROI to a various class based on a silhouette of motion difference and position of pedestrian-classified sub ROI. The system successfully categorizes five classes such as a single pedestrian, multiple pedestrians, a cyclist, a baby-car and a suitcase. As the process utilizes the output from pedestrian detection system, our system accomplishes categorization while computational costs hardly increased.

Language: ja

Keywords

Image processing; Pedestrian categorization; Pedestrian detection; Safety