Active Object Recognition via Monte Carlo Tree Search

Mikko Lauri; Nikolay Atanasov; George Pappas; Risto Ritala

Active Object Recognition via Monte Carlo Tree Search

Mikko Lauri, Nikolay Atanasov, George Pappas, Risto Ritala

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Professional

148 Downloads (Pure)

Abstract

This paper considers object recognition with a camera, whose viewpoint can be controlled in order to improve the recognition results. The goal is to choose a multi-view camera trajectory in order to minimize the probability of having misclassified objects and incorrect orientation estimates. Instead of using offline dynamic programming, the resulting stochastic optimal control problem is addressed via an online Monte Carlo tree search algorithm, which can handle various constraints and provides exceptional performance in large state spaces. A key insight is to use an active hypothesis testing policy to select camera viewpoints during the rollout stage of the tree search.

Original language	English
Title of host publication	ICRA 2015 Workshop: Beyond Geometric Constraints
Subtitle of host publication	Planning for Solving Complex Tasks, Reducing Uncertainty, and Generating Informative Paths & Policies
Number of pages	3
Publication status	Published - 30 May 2015
Publication type	D3 Professional conference proceedings

Keywords

Active classification
Object detection
Monte Carlo methods
Decision-making

Access to Document

Lauri_et_al_Active_ObjectFinal published version, 401 KB

Cite this

@inproceedings{445b36d4b584476f94ed8271da10cd9e,

title = "Active Object Recognition via Monte Carlo Tree Search",

abstract = "This paper considers object recognition with a camera, whose viewpoint can be controlled in order to improve the recognition results. The goal is to choose a multi-view camera trajectory in order to minimize the probability of having misclassified objects and incorrect orientation estimates. Instead of using offline dynamic programming, the resulting stochastic optimal control problem is addressed via an online Monte Carlo tree search algorithm, which can handle various constraints and provides exceptional performance in large state spaces. A key insight is to use an active hypothesis testing policy to select camera viewpoints during the rollout stage of the tree search.",

keywords = "Active classification, Object detection, Monte Carlo methods, Decision-making",

author = "Mikko Lauri and Nikolay Atanasov and George Pappas and Risto Ritala",

note = "xoa Lauri_et_al_Active_Object ei tarkistettu, siirretty kohdasta additional files ISBN kysytty, ei l{\"o}ydy / TL",

year = "2015",

month = may,

day = "30",

language = "English",

booktitle = "ICRA 2015 Workshop: Beyond Geometric Constraints",

}

Active Object Recognition via Monte Carlo Tree Search. / Lauri, Mikko; Atanasov, Nikolay; Pappas, George; Ritala, Risto.

ICRA 2015 Workshop: Beyond Geometric Constraints: Planning for Solving Complex Tasks, Reducing Uncertainty, and Generating Informative Paths & Policies. 2015. 8.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Professional

TY - GEN

T1 - Active Object Recognition via Monte Carlo Tree Search

AU - Lauri, Mikko

AU - Atanasov, Nikolay

AU - Pappas, George

AU - Ritala, Risto

N1 - xoa Lauri_et_al_Active_Object ei tarkistettu, siirretty kohdasta additional files ISBN kysytty, ei löydy / TL

PY - 2015/5/30

Y1 - 2015/5/30

N2 - This paper considers object recognition with a camera, whose viewpoint can be controlled in order to improve the recognition results. The goal is to choose a multi-view camera trajectory in order to minimize the probability of having misclassified objects and incorrect orientation estimates. Instead of using offline dynamic programming, the resulting stochastic optimal control problem is addressed via an online Monte Carlo tree search algorithm, which can handle various constraints and provides exceptional performance in large state spaces. A key insight is to use an active hypothesis testing policy to select camera viewpoints during the rollout stage of the tree search.

AB - This paper considers object recognition with a camera, whose viewpoint can be controlled in order to improve the recognition results. The goal is to choose a multi-view camera trajectory in order to minimize the probability of having misclassified objects and incorrect orientation estimates. Instead of using offline dynamic programming, the resulting stochastic optimal control problem is addressed via an online Monte Carlo tree search algorithm, which can handle various constraints and provides exceptional performance in large state spaces. A key insight is to use an active hypothesis testing policy to select camera viewpoints during the rollout stage of the tree search.

KW - Active classification

KW - Object detection

KW - Monte Carlo methods

KW - Decision-making

UR - http://people.csail.mit.edu/jingjin/ICRA15/

M3 - Conference contribution

BT - ICRA 2015 Workshop: Beyond Geometric Constraints

ER -

Active Object Recognition via Monte Carlo Tree Search

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this