Position estimation device, position estimation method, and program

• US 9,098,744 B2
• Filed: 02/10/2011
• Issued: 08/04/2015
• Est. Priority Date: 05/19/2010
• Status: Active Grant

• Current Assignee: Soinn Inc.
• Original Assignee: Tokyo Institute of Technology (National University Corporation Tokyo Institute of Technology)
• Inventors: Tongprasit, Noppharit, Kawewong, Aram, Hasegawa, Osamu
• Primary Examiner(s): Kassa, Yosef
• Application Number:
  US13/698,572
• Publication Number:
  US 20130108172A1
• Time in Patent Office: 1,636 Days
• Field of Search: 382/156, 382/157, 382/209, 382/218, 382/278, 382/291, 382/171, 382/190, 382/216, 382/282, 348/116, 348/161
• US Class Current: 1/1
• CPC Class Codes: G06T 7/73 using feature-based methods
• CPC Class Codes: G06V 20/10 Terrestrial scenes scenes under surveillance with static cameras G06V20/52; scenes perceived from the exterior of a vehicle G06V20/56; scenes perceived from the interior of a vehicle G06V20/59

Abstract

Disclosed is a position estimation device including a feature extraction unit that extracts invariant features from an input image, a matching unit that obtains matching between an input image and a registered place by referring to a database containing each registered place and invariant features in association, a similarity calculation unit that calculates a similarity with inclusion of a registered place near a selected registered place when the matching is a threshold or more, and a position identification unit that identifies the input image as a registered place when the similarity is a threshold or more. The feature extraction unit extracts local features from each input image being sequential images taken sequentially, selects features matched between the sequential images as sequential features, and calculates invariant features based on the sequential features. The number of sequential images is variable depending on the number of matched features.

Claims

1. A position estimation device comprising:

a feature extraction means for extracting invariant features from an input image;

a matching means for obtaining matching between an input image and a registered place in reference to a database containing each registered place and invariant features in association with each other;

a similarity calculation means for calculating a similarity with inclusion of a registered place near a selected registered place when the matching is equal to or larger than a specified threshold; and

a position identification means for identifying the input image as a registered place when the similarity is equal to or higher than a specified threshold, wherein

the feature extraction means includes

a local feature extraction means for extracting local features from each input image being sequential images taken sequentially,

a feature matching means for performing matching of the local features extracted by the local feature extraction means between the sequential input images based on matching scores of the sequential input images,

a sequential feature selection means for selecting the local features matched between the sequential images by the feature matching means as sequential features, and

an invariant feature calculation means for calculating invariant features based on the sequential features, and

the sequential feature selection means sets the number of sequential images to be variable depending on the number of matched features.

2. The position estimation device according to claim 1, wherein the matching means includes a common dictionary containing each features registered in association with an index and performs matching by substituting an index for local features of each input image in reference to the common dictionary.

3. The position estimation device according to claim 2, wherein the matching means calculates a matching score based on a product of the number of matches to features registered in the common dictionary and the number of matches to features contained in a matching target image.

4. The position estimation device according to claim 1, wherein

the similarity calculation means calculates a first estimated value by assigning a weight to a matching score of the selected registered place and the near registered place, and

the position identification means identifies whether the place is a registered place using the first estimated value as the similarity.

5. The position estimation device according to claim 1, wherein

the similarity calculation means calculates a second estimated value by normalizing the first estimated value, and

the position identification means identifies whether the place is a registered place using the second estimated value as the similarity.

6. The position estimation device according to claim 1, wherein the local features are SIFT (Scale Invariant Feature Transformation) and/or SURF (Speed Up Robustness Features).

7. A position estimation method comprising:

a feature extraction step of extracting invariant features from an input image;

a matching step of obtaining matching between an input image and a registered place in reference to a database containing each registered place and invariant features in association with each other;

a similarity calculation step of calculating a similarity with inclusion of a registered place near a selected registered place when the matching is equal to or larger than a specified threshold; and

a position identification step of identifying the input image as a registered place when the similarity is equal to or higher than a specified threshold, wherein

the feature extraction step includes

a local feature extraction step of extracting local features from each input image being sequential images taken sequentially,

a feature matching step of performing matching of the local features extracted in the local feature extraction step between the sequential input images based on matching scores of the sequential input images,

a sequential feature selection step of selecting the local features matched between the sequential images in the feature matching step as sequential features, and

an invariant feature calculation step of calculating invariant features based on the sequential features, and

in the sequential feature selection step, the number of sequential images is variable depending on the number of matched features.

8. The position estimation method according to claim 7, wherein the matching step performs matching by substituting an index for local features of each input image in reference to a common dictionary containing each feature registered in association with an index.

9. The position estimation method according to claim 8, wherein the matching step calculates a matching score based on a product of the number of matches to features registered in the common dictionary and the number of matches to features contained in a matching target image.

10. The position estimation method according to claim 7, wherein

the similarity calculation step calculates a first estimated value by assigning a weight to a matching score of the selected registered place and the near registered place, and

the identification step identifies whether the place is a registered place using the first estimated value as the similarity.

11. The position estimation method according to claim 7, wherein

the similarity calculation step calculates a second estimated value by normalizing the first estimated value, and

the identification step identifies whether the place is a registered place using the second estimated value as the similarity.

12. The position estimation method according to claim 7, wherein the local features are SIFT (Scale Invariant Feature Transformation) and/or SURF (Speed Up Robustness Features).

13. A program on a non-transitory computer readable medium causing a computer to execute a specified operation comprising:

a feature extraction step of extracting invariant features from an input image;

a matching step of obtaining matching between an input image and a registered place in reference to a database containing each registered place and invariant features in association with each other;

a similarity calculation step of calculating a similarity with inclusion of a registered place near a selected registered place when the matching is equal to or larger than a specified threshold; and

a position identification step of identifying the input image as a registered place when the similarity is equal to or higher than a specified threshold, wherein

the feature extraction step includes

a local feature extraction step of extracting local features from each input image being sequential images taken sequentially,

a feature matching step of performing matching of the local features extracted in the local feature extraction step between the sequential input images based on matching scores of the sequential input images,

a sequential feature selection step of selecting the local features matched between the sequential images in the feature matching step as sequential features, and

an invariant feature calculation step of calculating invariant features based on the sequential features, and

in the sequential feature selection step, the number of sequential images is variable depending on the number of matched features.