Details
| Originalsprache | Englisch |
|---|---|
| Titel des Sammelwerks | 38. Wissenschaftlich-Technische Jahrestagung der DGPF und PFGK18 Tagung in München |
| Untertitel | Publikationen der DGPF |
| Seitenumfang | 13 |
| Band | 27 |
| Publikationsstatus | Veröffentlicht - 2018 |
| Veranstaltung | PFGK 18: Photogrammetrie-Fernerkundung-Geoinformatik-Kartografie-2018 - Technische Universität München, München, Deutschland Dauer: 7 März 2018 → 9 März 2018 |
Publikationsreihe
| Name | Publikationen der Deutschen Gesellschaft für Photogrammetrie, Fernerkundung und Geoinformation e.V. |
|---|---|
| Band | 27 |
| ISSN (Print) | 0942-2870 |
Abstract
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
38. Wissenschaftlich-Technische Jahrestagung der DGPF und PFGK18 Tagung in München: Publikationen der DGPF. Band 27 2018. (Publikationen der Deutschen Gesellschaft für Photogrammetrie, Fernerkundung und Geoinformation e.V.; Band 27).
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung › Peer-Review
}
TY - GEN
T1 - Object Proposals for Pedestrian Detection in Stereo Images
AU - Nguyen, Uyen
AU - Rottensteiner, Franz
AU - Heipke, Christian
PY - 2018
Y1 - 2018
N2 - Pedestrian detection is an active research field in computer vision and photogrammetry today due to its importance for applications related to autonomous driving, machine human interaction, and surveillance. Object proposals play a significant role in guiding a classifier where to examine image regions that may contain pedestrians. In this paper, we present a proposal framework employing both 3D information derived from stereo images and RGB cues to generate pedestrian bounding box proposals with high recall and relatively small number of candidates. Our proposal generator has two stages: (1) generating a large number of proposals to achieve a good recall value; and (2) eliminating unreasonable initial candidates based on 3D cues and pedestrian geometric constraints. Preliminary experiments on the Kitti benchmark dataset show that our proposal framework is comparable to state-of-the-art methods.
AB - Pedestrian detection is an active research field in computer vision and photogrammetry today due to its importance for applications related to autonomous driving, machine human interaction, and surveillance. Object proposals play a significant role in guiding a classifier where to examine image regions that may contain pedestrians. In this paper, we present a proposal framework employing both 3D information derived from stereo images and RGB cues to generate pedestrian bounding box proposals with high recall and relatively small number of candidates. Our proposal generator has two stages: (1) generating a large number of proposals to achieve a good recall value; and (2) eliminating unreasonable initial candidates based on 3D cues and pedestrian geometric constraints. Preliminary experiments on the Kitti benchmark dataset show that our proposal framework is comparable to state-of-the-art methods.
M3 - Conference contribution
VL - 27
T3 - Publikationen der Deutschen Gesellschaft für Photogrammetrie, Fernerkundung und Geoinformation e.V.
BT - 38. Wissenschaftlich-Technische Jahrestagung der DGPF und PFGK18 Tagung in München
T2 - PFGK 18
Y2 - 7 March 2018 through 9 March 2018
ER -