Object detection method, information processing device, and storage medium

Patent metadata

Field	Value
Publication number	US-9734416-B2
Application number	US-201615018002-A
Country	US
Kind code	B2
Filing date	Feb 8, 2016
Priority date	Mar 23, 2015
Publication date	Aug 15, 2017
Grant date	Aug 15, 2017

Abstract

An object detection method includes extracting an object candidate by grouping a plurality of characteristic points from two images acquired at different times by a camera; calculating, for each of characteristic points included in the object candidate, an observation flow amount indicating an amount of positional shift on the two images and a virtual road surface flow amount indicating an amount of positional shift on the two images while assuming that the characteristic point is located on a road surface; generating, for each of the characteristic points, a flow difference distribution of the object candidate by calculating a difference between the observation flow amount and the virtual road surface flow amount; and determining whether the object candidate is an object existing on and protruding from the road surface by comparing the flow difference distribution with a plurality of flow difference distribution models generated for predicted inclined states of the surface.

First claim

What is claimed is:

1. An object detection method executed by a processor included in an object detecting device having a camera attached to a moving object, the object detection method comprising:
extracting a plurality of characteristic points from two images acquired at different times by the camera upon a movement of the moving object;
extracting an object candidate by grouping the plurality of characteristic points;
calculating, for each of characteristic points included in the object candidate, an observation flow amount indicating an amount of positional shift of a characteristic point on the two images;
calculating, for each of the characteristic points included in the object candidate, a virtual road surface flow amount indicating an amount of positional shift of the characteristic point on the two images while assuming that the characteristic point is located on a road surface;
generating, for each of the characteristic points included in the object candidate, a flow difference distribution of the object candidate by calculating a difference between the observation flow amount and the virtual road surface flow amount; and
determining whether the object candidate is an object existing on and protruding from the road surface by comparing the flow difference distribution with a plurality of flow difference distribution models generated for predicted inclined states of the surface and corresponding to at least any of a case where the object candidate is the object existing along the road surface and a case where the object candidate is the object existing on and protruding from the road surface.

2. The object detection method according to claim 1 ,
wherein the determining includes
identifying, based on similarities between the plurality of flow difference distribution models and the generated flow difference distribution, a flow difference distribution model corresponding to the object candidate from among the plurality of flow difference distribution models, and
determining that the object candidate is the object existing on and protruding from the surface when it is determined that the flow difference distribution model is related to the object existing on and protruding from the road surface.

3. The object detection method according to claim 2 ,
wherein the determining includes
selecting, from among the plurality of flow difference distribution models, a flow difference distribution model of which a deviation from the generated flow difference distribution is the smallest, and
determining the selected flow difference distribution model as a flow difference distribution model corresponding to the object candidate when the deviation of the selected flow difference distribution model is smaller than a predetermined threshold.

4. The object detection method according to claim 3 ,
wherein the determining includes
separating the flow difference distribution into a plurality of flow difference distributions when the deviation of the selected flow difference distribution model is larger than the predetermined threshold, and
comparing the plurality of flow difference distributions with the plurality of flow difference distribution models and thereby determining whether each of images corresponding to the plurality of flow difference distributions are images of the object existing on and protruding from the road surface.

5. The object detection method according to claim 4 ,
wherein the separating includes
acquiring the deviation at predetermined time intervals while increasing a coordinate value in a predetermined coordinate axis direction, and
separating the flow difference distribution into two flow difference distributions in a direction perpendicular to the predetermined coordinate axis direction at a position at which the deviation changes from a level equal to or smaller than the predetermined threshold to a level higher than the predetermined threshold.

6. The object detection method according to claim 1 , further comprising:
identifying a flow difference distribution model corresponding to a first object from among the plurality of flow difference distribution models when a plurality of object candidates including the first object and a second object is extracted from the plurality of images;
determining whether the first object is the object existing on and protruding from the road surface;
identifying a flow difference distribution model corresponding to the second object from among the plurality of flow difference distribution models; and
determining whether the second object is the object existing on and protruding from the road surface.

7. The object detection method according to claim 1 ,
wherein the object that exists on the road surface while protruding from the surface is a cubic object and the object that exists along the road surface is a road surface pattern.

8. The object detection method according to claim 1 ,
wherein the plurality of flow difference distribution models is generated for combinations of object types and the inclined states of the surface.

9. An information processing device comprising:
a camera; and
a processor coupled to the camera and configured to
extract a plurality of characteristic points from two images acquired at different times by the camera upon a movement of the moving object,
extract an object candidate by grouping the plurality of characteristic points,
calculate, for each of characteristic points included in the object candidate, an observation flow amount indicating an amount of positional shift of a characteristic point on the two images,
calculate, for each of the characteristic points included in the object candidate, a virtual road surface flow amount indicating an amount of positional shift of the characteristic point on the two images while assuming that the characteristic point is located on a road surface,
generate, for each of the characteristic points included in the object candidate, a flow difference distribution of the object candidate by calculating a difference between the observation flow amount and the virtual road surface flow amount, and
determine whether the object candidate is an object existing on and protruding from the road surface by comparing the flow difference distribution with a plurality of flow difference distribution models generated for predicted inclined states of the surface and corresponding to at least any of a case where the object candidate is the object existing along the road surface and a case where the object candidate is the object existing on and protruding from the road surface.

10. A non-transitory computer-readable storage medium storing a program that causes a processor included in a computer to execute a process, the process comprising:
extracting a plurality of characteristic points from two images acquired at different times by the camera upon a movement of the moving object;
extracting an object candidate by grouping the plurality of characteristic points;
calculating, for each of the plurality of characteristic points, an observation flow amount indicating an amount of positional shift of a characteristic point on the two images;
calculating, for each of the plurality of characteristic points, a virtual road surface flow amount indicating an amount of positional shift of the characteristic point on the two images while assuming that the characteristic point is located on a road surface;
generating, for each of the plurality of characteristic points, a flow difference distribution of the object candidate by calculating a difference between the observation flow amount and the virtual road surface flow amount; and
determining whether the object candidate is an o

Assignees

• Fujitsu Ltd

Inventors

• Kanetake Jun
• Nakayama Osafumi

Classifications

• G06T2207/30241

  Trajectory · CPC title

• G06T2207/30261

  Obstacle · CPC title

• G06T2207/10016

  Video; Image sequence · CPC title

• G06T7/269 Primary

  using gradient-based methods · CPC title

• G06K9/00805 Primary

  Physics · mapped topic