https://sites.google.com/site/pedestrianbenchmarkhttp://rcv.kaist.ac.kr/multispectral-pedestrian/As increasing of interest in pedestrian detection, the dataset has been also subject to the research in the past decades. Two of the most noticeable are needs for much closer real traffic conditions and for meaningful information for detection. Although color information is very useful, sometimes it is limited to detect pedestrian. In these respects, we propose a new multispectral pedestrian detection dataset which provides aligned color-thermal image pairs of the general traffic scenes using beam splitter. With the new dataset, we extend aggregated channel features (ACF) algorithm to handle additional information from aligned thermal images. From our extension of ACF, we reduce averaging miss rate by 15% on the proposed dataset. We expect this multispectral dataset to make a breakthrough in present pedestrian detections.
[Acknowledgement]
We thank anonymous reviewers giving constructive comments to our work. We also appreciate KAIST-RCV labmates who help to finish the tedious annotation task.
This work was supported by the Development of Autonomous Emergency Braking System for Pedestrian Protection project funded by the Ministry of Trade, Industry and Energy of Korea. (MOTIE)(No.10044775)
[Acknowledgement]
We thank anonymous reviewers giving constructive comments to our work. We also appreciate KAIST-RCV labmates who help to finish the tedious annotation task.
This work was supported by the Development of Autonomous Emergency Braking System for Pedestrian Protection project funded by the Ministry of Trade, Industry and Energy of Korea. (MOTIE)(No.10044775)