| Item type |
SIG Technical Reports(1) |
| 公開日 |
2023-05-11 |
| タイトル |
|
|
タイトル |
View Birdification: On-Ground Pedestrian Movement Estimation and Prediction from Ego-centric In-Crowd Views |
| タイトル |
|
|
言語 |
en |
|
タイトル |
View Birdification: On-Ground Pedestrian Movement Estimation and Prediction from Ego-centric In-Crowd Views |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
D論セッション |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_18gh |
|
資源タイプ |
technical report |
| 著者所属 |
|
|
|
OMRON SINIC X／Graduate School of Informatics, Kyoto University |
| 著者所属 |
|
|
|
Graduate School of Informatics, Kyoto University |
| 著者所属 |
|
|
|
Graduate School of Informatics, Kyoto University |
| 著者所属(英) |
|
|
|
en |
|
|
OMRON SINIC X / Graduate School of Informatics, Kyoto University |
| 著者所属(英) |
|
|
|
en |
|
|
Graduate School of Informatics, Kyoto University |
| 著者所属(英) |
|
|
|
en |
|
|
Graduate School of Informatics, Kyoto University |
| 著者名 |
Mai, Nishimura
Shohei, Nobuhara
Ko, Nishino
|
| 著者名(英) |
Mai, Nishimura
Shohei, Nobuhara
Ko, Nishino
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
This paper introduces view birdification, a task of recovering ground-plane movements of people in a crowd from an ego-centric video captured by an observer also moving in the crowd. Unlike conventional geometric reconstruction methods that assume a static world, view birdification relies solely on the perceived movements of dynamic objects. The key difficulty of this task is that the two kinds of trajectories, the camera ego-motion and pedestrian trajectories, are deeply intertwined in the observed movements. To address this, we present a cascaded optimization approach from a Bayesian perspective and a data-driven solver for view birdification which simultaneously learn an underlying motion model. Furthermore, we extend view birdification as an object-oriented world model that can estimate the future state of each pedestrian on the ground solely from the ego-centric view observation. Our extensive evaluation demonstrates the effectiveness of our methods in diverse densities of crowds, and shows some promising results in zero-shot adaptation to real video sequences. We believe view birdification will serve as a sound foundation for crowd and pedestrian movement modeling and enable a wide range of downstream applications including but not limited to navigation. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
This paper introduces view birdification, a task of recovering ground-plane movements of people in a crowd from an ego-centric video captured by an observer also moving in the crowd. Unlike conventional geometric reconstruction methods that assume a static world, view birdification relies solely on the perceived movements of dynamic objects. The key difficulty of this task is that the two kinds of trajectories, the camera ego-motion and pedestrian trajectories, are deeply intertwined in the observed movements. To address this, we present a cascaded optimization approach from a Bayesian perspective and a data-driven solver for view birdification which simultaneously learn an underlying motion model. Furthermore, we extend view birdification as an object-oriented world model that can estimate the future state of each pedestrian on the ground solely from the ego-centric view observation. Our extensive evaluation demonstrates the effectiveness of our methods in diverse densities of crowds, and shows some promising results in zero-shot adaptation to real video sequences. We believe view birdification will serve as a sound foundation for crowd and pedestrian movement modeling and enable a wide range of downstream applications including but not limited to navigation. |
| 書誌レコードID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AA11131797 |
| 書誌情報 |
研究報告コンピュータビジョンとイメージメディア（CVIM）
巻 2023-CVIM-234,
号 3,
p. 1-16,
発行日 2023-05-11
|
| ISSN |
|
|
収録物識別子タイプ |
ISSN |
|
収録物識別子 |
2188-8701 |
| Notice |
|
|
|
SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. |
| 出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |
IPSJ-CVIM23234003.pdf (6.7 MB)
