On recall rate of interest point detectors

Henrik Aanæs; Anders Lindbjerg Dahl; Kim Steenstrup Pedersen

On recall rate of interest point detectors

Henrik Aanæs, Anders Lindbjerg Dahl, Kim Steenstrup Pedersen

Department of Computer Science

6 Downloads (Pure)

Abstract

In this paper we provide a method for evaluating interest point detectors independently of image descriptors. This is possible because we have compiled a unique data set enabling us to determine if common interest points are found. The data contains 60 scenes of a wide range of object types, and for each scene we have 119 precisely located camera positions obtained from a camera mounted on an industrial robot arm. The scene surfaces have been scanned using structured light, providing precise 3D
ground truth. We have investigated a number of the most popular interest point detectors where we systematically have varied camera position, light and model parameters. The overall conclusion is that the Harris and Hessian corner detectors perform well followed by MSER, whereas the FAST corner
detector, IBR and EBR performs poorly. Furthermore, only the number of interest points change with changing parameters - not the correct matches.

Original language	English
Title of host publication	Electronic Proceedings of 3DPVT'10 : The Fifth International Symposium on 3D Data Processing, Visualization and Transmission
Number of pages	8
Publication date	2010
Pages	1-8
Publication status	Published - 2010
Event	5th International Symposium 3D Data Processing, Visualization and Transmission - Paris, France Duration: 17 May 2010 → 20 May 2010 Conference number: 5

Conference

Conference	5th International Symposium 3D Data Processing, Visualization and Transmission
Number	5
Country/Territory	France
City	Paris
Period	17/05/2010 → 20/05/2010

Access to Document

Aanæs_2010_On_recall_rateAccepted author manuscript, 1.17 MB

Cite this

@inproceedings{8cbf7e903cc811df928f000ea68e967b,

title = "On recall rate of interest point detectors",

abstract = "In this paper we provide a method for evaluating interest point detectors independently of image descriptors. This is possible because we have compiled a unique data set enabling us to determine if common interest points are found. The data contains 60 scenes of a wide range of object types, and for each scene we have 119 precisely located camera positions obtained from a camera mounted on an industrial robot arm. The scene surfaces have been scanned using structured light, providing precise 3Dground truth. We have investigated a number of the most popular interest point detectors where we systematically have varied camera position, light and model parameters. The overall conclusion is that the Harris and Hessian corner detectors perform well followed by MSER, whereas the FAST cornerdetector, IBR and EBR performs poorly. Furthermore, only the number of interest points change with changing parameters - not the correct matches. ",

author = "Henrik Aan{\ae}s and {Lindbjerg Dahl}, Anders and Pedersen, {Kim Steenstrup}",

note = "Best paper award.; 5th International Symposium 3D Data Processing, Visualization and Transmission, 3DPVT 2010 ; Conference date: 17-05-2010 Through 20-05-2010",

year = "2010",

language = "English",

pages = "1--8",

booktitle = "Electronic Proceedings of 3DPVT'10",

}

TY - GEN

T1 - On recall rate of interest point detectors

AU - Aanæs, Henrik

AU - Lindbjerg Dahl, Anders

AU - Pedersen, Kim Steenstrup

N1 - Conference code: 5

PY - 2010

Y1 - 2010

N2 - In this paper we provide a method for evaluating interest point detectors independently of image descriptors. This is possible because we have compiled a unique data set enabling us to determine if common interest points are found. The data contains 60 scenes of a wide range of object types, and for each scene we have 119 precisely located camera positions obtained from a camera mounted on an industrial robot arm. The scene surfaces have been scanned using structured light, providing precise 3Dground truth. We have investigated a number of the most popular interest point detectors where we systematically have varied camera position, light and model parameters. The overall conclusion is that the Harris and Hessian corner detectors perform well followed by MSER, whereas the FAST cornerdetector, IBR and EBR performs poorly. Furthermore, only the number of interest points change with changing parameters - not the correct matches.

AB - In this paper we provide a method for evaluating interest point detectors independently of image descriptors. This is possible because we have compiled a unique data set enabling us to determine if common interest points are found. The data contains 60 scenes of a wide range of object types, and for each scene we have 119 precisely located camera positions obtained from a camera mounted on an industrial robot arm. The scene surfaces have been scanned using structured light, providing precise 3Dground truth. We have investigated a number of the most popular interest point detectors where we systematically have varied camera position, light and model parameters. The overall conclusion is that the Harris and Hessian corner detectors perform well followed by MSER, whereas the FAST cornerdetector, IBR and EBR performs poorly. Furthermore, only the number of interest points change with changing parameters - not the correct matches.

M3 - Article in proceedings

SP - 1

EP - 8

BT - Electronic Proceedings of 3DPVT'10

T2 - 5th International Symposium 3D Data Processing, Visualization and Transmission

Y2 - 17 May 2010 through 20 May 2010

ER -

On recall rate of interest point detectors

Abstract

Conference

Access to Document

Fingerprint

Cite this