Deep Reinforcement Learning for Active Human Pose Estimation
(2020) 34th AAAI Conference on Artificial Intelligence, AAAI 2020 In Proceedings of the AAAI Conference on Artificial Intelligence 34(07). p.10835-10844- Abstract
- Most 3d human pose estimation methods assume that input – be it images of a scene collected from one or several viewpoints, or from a video – is given. Consequently, they focus on estimates leveraging prior knowledge and measurement by fusing information spatially and/or temporally, whenever available. In this paper we address the problem of an active observer with freedom to move and explore the scene spatially – in ‘time-freeze’ mode – and/or temporally, by selecting informative viewpoints that improve its estimation accuracy. Towards this end, we introduce Pose-DRL, a fully trainable deep reinforcement learning-based active pose estimation architecture which learns to select appropriate views, in space and time, to feed an underlying... (More)
- Most 3d human pose estimation methods assume that input – be it images of a scene collected from one or several viewpoints, or from a video – is given. Consequently, they focus on estimates leveraging prior knowledge and measurement by fusing information spatially and/or temporally, whenever available. In this paper we address the problem of an active observer with freedom to move and explore the scene spatially – in ‘time-freeze’ mode – and/or temporally, by selecting informative viewpoints that improve its estimation accuracy. Towards this end, we introduce Pose-DRL, a fully trainable deep reinforcement learning-based active pose estimation architecture which learns to select appropriate views, in space and time, to feed an underlying monocular pose estimator. We evaluate our model using single- and multi-target estimators with strong result in both settings. Our system further learns automatic stopping conditions in time and transition functions to the next temporal processing step in videos. In extensive experiments with the Panoptic multi-view setup, and for complex scenes containing multiple people, we show that our model learns to select viewpoints that yield significantly more accurate pose estimates compared to strong multi-view baselines. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/58668314-e07a-4641-8480-8c4ad50c8450
- author
- Gärtner, Erik LU ; Pirinen, Aleksis LU and Sminchisescu, Cristian LU
- organization
- publishing date
- 2020-04-03
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- AAAI 2020 - 34th AAAI Conference on Artificial Intelligence
- series title
- Proceedings of the AAAI Conference on Artificial Intelligence
- volume
- 34
- issue
- 07
- pages
- 10835 - 10844
- publisher
- The Association for the Advancement of Artificial Intelligence
- conference name
- 34th AAAI Conference on Artificial Intelligence, AAAI 2020
- conference location
- New York, United States
- conference dates
- 2020-02-07 - 2020-02-12
- external identifiers
-
- scopus:85095322803
- ISSN
- 2159-5399
- DOI
- 10.1609/aaai.v34i07.6714
- project
- Deep Learning for Understanding Humans
- language
- English
- LU publication?
- yes
- id
- 58668314-e07a-4641-8480-8c4ad50c8450
- date added to LUP
- 2021-04-08 10:52:40
- date last changed
- 2024-01-13 15:31:37
@inproceedings{58668314-e07a-4641-8480-8c4ad50c8450, abstract = {{Most 3d human pose estimation methods assume that input – be it images of a scene collected from one or several viewpoints, or from a video – is given. Consequently, they focus on estimates leveraging prior knowledge and measurement by fusing information spatially and/or temporally, whenever available. In this paper we address the problem of an active observer with freedom to move and explore the scene spatially – in ‘time-freeze’ mode – and/or temporally, by selecting informative viewpoints that improve its estimation accuracy. Towards this end, we introduce Pose-DRL, a fully trainable deep reinforcement learning-based active pose estimation architecture which learns to select appropriate views, in space and time, to feed an underlying monocular pose estimator. We evaluate our model using single- and multi-target estimators with strong result in both settings. Our system further learns automatic stopping conditions in time and transition functions to the next temporal processing step in videos. In extensive experiments with the Panoptic multi-view setup, and for complex scenes containing multiple people, we show that our model learns to select viewpoints that yield significantly more accurate pose estimates compared to strong multi-view baselines.}}, author = {{Gärtner, Erik and Pirinen, Aleksis and Sminchisescu, Cristian}}, booktitle = {{AAAI 2020 - 34th AAAI Conference on Artificial Intelligence}}, issn = {{2159-5399}}, language = {{eng}}, month = {{04}}, number = {{07}}, pages = {{10835--10844}}, publisher = {{The Association for the Advancement of Artificial Intelligence}}, series = {{Proceedings of the AAAI Conference on Artificial Intelligence}}, title = {{Deep Reinforcement Learning for Active Human Pose Estimation}}, url = {{http://dx.doi.org/10.1609/aaai.v34i07.6714}}, doi = {{10.1609/aaai.v34i07.6714}}, volume = {{34}}, year = {{2020}}, }