MAP GENERATION APPARATUS, MAP GENERATION METHOD, MOVING METHOD FOR MOVING BODY, AND ROBOT APPARATUS

• US 20130216098A1
• Filed: 08/30/2011
• Published: 08/22/2013
• Est. Priority Date: 09/17/2010
• Status: Active Grant

• Current Assignee: Tokyo Institute of Technology (National University Corporation Tokyo Institute of Technology)
• Original Assignee: Tokyo Institute of Technology (National University Corporation Tokyo Institute of Technology)
• Inventors: Hasegawa, Osamu, Morioka, Hiroshi, Tongprasit, Noppharit, Yi, Sangkyu
• Granted Patent: US 9,224,043 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G05D 1/0253 extracting relative motion information from a plurality of images taken successively, e.g. visual odometry, optical flow determining position or orientation from images G06T7/70
• CPC Class Codes: G05D 1/0274 using mapping information stored in a memory device navigation using map-matching G01C21/30
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/30241 Trajectory
• CPC Class Codes: G06T 2207/30244 Camera pose
• CPC Class Codes: G06T 7/246 using feature-based methods, e.g. the tracking of corners or segments
• CPC Class Codes: G06T 7/579 from motion
• 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
• CPC Class Codes: Y10S 901/01 Mobile robot

Abstract

Performing map construction under a crowded environment where there are a lot of people. It includes a successive image acquisition unit that obtains images that are taken while a robot is moving, a local feature quantity extraction unit that extracts a quantity at each feature point from the images, a feature quantity matching unit that performs matching among the quantities in the input images, where quantities are extracted by the extraction unit, an invariant feature quantity calculation unit that calculates an average of the matched quantities among a predetermined number of images by the matching unit as an invariant feature quantity, a distance information acquisition unit that calculates distance information corresponding to each invariant feature quantity based on a position of the robot at times when the images are obtained, and a map generation unit that generates a local metrical map as a hybrid map.

Claims

1. A map generation method comprising:

obtaining successive images that are successively taken while a moving body is moving;

extracting a local feature quantity at each feature point from the successive images obtained in the successive image acquisition step;

performing matching among the local feature quantities in the successive input images, the local feature quantity being extracted in the local feature quantity extraction step;

calculating an average of matched local feature quantities among a predetermined number of successive images in the feature quantity matching step as an invariant feature quantity;

calculating distance information corresponding to each invariant feature quantity based on a position of the moving body at each point in time when the successive images are obtained; and

generating a local metrical map as a hybrid map including the invariant feature quantity and the distance information.

2. The map generation method according to claim 1, further comprising obtaining a position and a posture of the moving body, and a position of a feature point corresponding to the invariant feature quantity from information and an observation values, the information indicating about a movement of the moving body at each point in time when a series of a plurality of successive images are obtained, the observation value indicating a measurement direction of a feature point having the invariant feature quantity, and the measurement direction being seen from the moving body, wherein

in the map generation step, the hybrid map including a position and a posture of the moving body, and a position and an invariant feature quantity of a feature point are generated.

3. The map generation method according to claim 2, wherein when acquiring position information a stochastically most likely position and posture of the moving body and a position of the feature point are calculated from a group of information and a group of the observation values, the group of the information indicating about the moving body at each point in time when the plurality of successive images from which the invariant feature quantity are obtained.

4. The map generation method according to claim 2, wherein when acquiring position information the hybrid map including a position and a posture of the moving body at each point in time and a position of the feature point are generated, the position and the posture of the moving body and the position of the feature point being obtained among the plurality of successive images by minimizing a below-shown expression with an observation value z_ij and an input a the observation value z_ij indicating in which direction the invariant feature quantity is seen from the moving body at each point in time, and the input u_i indicating distance information that is used when the moving body moves from x_j to x_i-1, when the position and the posture of the moving body being represented by x_i

5. The map generation method according to claim 2, wherein

the invariant feature quantity and a position and a posture of the moving body are obtained from a plurality of successive images, and

the generation of the map includes deleting past successive images so that the number of successive images that are used to calculate the position and the posture of the moving body and the invariant feature quantity is equal at each point in time, and deleting the invariant feature quantity that disappears as a result of deleting the successive images.

6. The map generation method according to claim 2, further comprising:

determining whether or not a current place was visited in past; and

correcting a previous position and posture x_Li0 of the moving body and a current position and posture x_Li the moving body based on x_Li0 and x_Li when the current place was visited in past.

7. The map generation method according to claim 6, wherein the values x_Li0 and x_Li are corrected by minimizing a below-shown expression:

8. A map generation apparatus comprising:

successive image acquisition means for Obtaining successive images that are successively taken while a moving body is moving;

local feature quantity extraction means for extracting a local feature quantity at each feature point from the successive images obtained by the successive image acquisition means;

feature quantity matching means for performing matching among the local feature quantities in the successive input images extracted by the local feature quantity extraction means;

invariant feature quantity calculation means for calculating an average of matched local feature quantities among a predetermined number of successive images by the feature quantity matching means as an invariant feature quantity;

distance information acquisition means for calculating distance information corresponding to each invariant feature quantity based on a position of the moving body at each point in time when the successive images are obtained; and

map generation means for generating a local metrical map as a hybrid map including the invariant feature quantity and the distance information.

9. The map generation apparatus according to claim 8, further comprising a position information acquisition means for obtaining a position and a posture of the moving body, and a position of a feature point corresponding to the invariant feature quantity from information and an observation value, the information indicating about a movement of the moving body at each point in time when a series of a plurality of successive images are obtained, the observation value indicating a measurement direction of a feature point having the invariant feature quantity, and the measurement direction being seen from the moving body, wherein

the map generation means generates the hybrid map including a position and a posture of the moving body, and a position and an invariant feature quantity of a feature point are generated.

10. The map generation apparatus according to claim 9, wherein the position information acquisition means calculates a stochastically most likely position and posture of the moving body, and a position of the feature point calculated from a group of information and a group of the observation values, the group of the information indicating about the moving body at each point in time when the plurality of successive images from which the invariant feature quantity are obtained.

11. A moving method for a moving body that performs route planning according to a local metrical map created in advance and thereby moves, comprising:

obtaining successive images while a moving body is moving;

extracting a local feature quantity at each feature point from ^.the successive images obtained in the successive image acquisition step;

performing matching among the local feature quantities in the successive input images, the local feature quantities being extracted in the local feature quantity extraction step;

calculating an average of matched local feature quantities among a predetermined number of successive images in the feature quantity matching step as an invariant feature quantity;

calculating distance information corresponding to each invariant feature quantity based on a position of the moving body at each point in time when the successive images are obtained; and

comparing the calculated invariant feature quantity and the distance information with an invariant feature quantity and distance information registered on the local metrical map, determining whether or not a current place is registered, and when the current place is registered, correcting the current place based on information of the local metrical map, wherein

the local metrical map is a hybrid map including distance information corresponding to each invariant feature quantity calculated based on position information at each point in time when the moving body has obtained the successive images and including the invariant feature quantity.

12. The moving method for a moving body according to claim 11, wherein when correcting the current line based on information of the local metrical map, when a current place is registered on the local metrical map, a registered position and posture x_Learnedi0 of the moving body and a current position and posture x_Li of the moving body are corrected based on x_Learnedi0 and x_Li.

13. The moving method for a moving body according to claim 12, wherein when correcting the current place based on information of the local metrical map, the values x_Learnedi0 and x_Li are corrected by minimizing a below-shown expression:

14. The moving method for a moving body according to claim 13, further comprising calculating a relative coordinate indicating how much a current place is different from a place registered on a map, wherein

when correcting the current place based on information of the local metrical map, the locus is corrected by using the relative coordinate as the value x_Li.

15. A robot apparatus that performs route planning according to a local metrical map created in advance and thereby moves, comprising:

successive image acquisition means for obtaining successive images while a moving body is moving;

local feature quantity extraction means for extracting a local feature quantity at each feature point from the successive images obtained by the successive image acquisition means;

feature quantity matching means for performing matching among the local feature quantities in the successive input images extracted by the local feature quantity extraction means;

invariant feature quantity calculation means for calculating an average of matched local feature quantities among a predetermined number of successive images, the matched local feature quantities being obtained by the feature quantity matching means as an invariant feature quantity;

distance information acquisition means for calculating distance information corresponding to each invariant feature quantity based on a position of the moving body at each point in time when the successive images are obtained; and

locus correction means for comparing the calculated invariant feature quantity and the distance information with an invariant feature quantity and distance information registered on the local metrical map, determining whether or not a current place is registered, and when the current place is registered, correcting the current place based on information of the local metrical map, wherein

the local metrical map is a hybrid map including distance information corresponding to each invariant feature quantity calculated based on position information at each point in time when the moving body has obtained the successive images and including the invariant feature quantity.