Road obstacle detection device, method, and program

Patent metadata

Field	Value
Publication number	US-2018330615-A1
Application number	US-201815964748-A
Country	US
Kind code	A1
Filing date	Apr 27, 2018
Priority date	May 12, 2017
Publication date	Nov 15, 2018
Grant date	—

Abstract

Provided is a road obstacle detection method including: a region segmentation step of segmenting an input image into a plurality of local regions; and a likelihood calculation step of calculating a probability of presence of the road obstacle in a target local region based on a probability that the target local region is not a normal object set in advance and a degree of visual saliency defined by a relationship between a surrounding local region and the target local region, wherein the degree of the visual saliency is calculated to be larger as a probability that the surrounding local region is a road is higher and calculated to be larger as a difference in visual characteristics between the target local region and the surrounding local region is larger.

First claim

What is claimed is:

1 . A road obstacle detection device that detects a road obstacle in an image, the road obstacle detection device comprising:
an image input unit configured to receive an input of an image; a region segmentation unit configured to segment the image into a plurality of local regions; and a likelihood calculation unit configured to calculate a probability of presence of the road obstacle in a target local region based on a probability that the target local region is not a normal object set in advance and a degree of visual saliency defined by a relationship between a surrounding local region and the target local region, wherein the degree of the visual saliency is calculated to be larger as a probability that the surrounding local region is a road is higher and calculated to be larger as a difference in visual characteristics between the target local region and the surrounding local region is larger.

2 . The road obstacle detection device according to claim 1 , wherein the degree of the visual saliency is calculated to be larger as a size of the surrounding local region is larger.

3 . The road obstacle detection device according to claim 1 , wherein the degree of the visual saliency is calculated to be larger as a distance between the target local region and the surrounding local region is smaller.

4 . The road obstacle detection device according to claim 1 , wherein the likelihood calculation unit calculates the probability of the presence of the road obstacle in the target local region as a product of a sum of the degrees of the visual saliency of the plurality of local regions and the probability that the target local region is not the normal object.

5 . The road obstacle detection device according to claim 1 , further comprising a discrimination unit configured to learn a plurality of normal objects in advance and discriminate the plurality of normal objects,
wherein a probability that the target local region is an object other than the normal object set in advance is determined based on a discrimination result by the discrimination unit.

6 . The road obstacle detection device according to claim 5 ,
wherein the discrimination unit discriminates the normal objects for each pixel, and wherein the probability that the target local region is the object other than the normal object set in advance is calculated as an average of probabilities that the target local region is not the normal object, the probabilities being calculated for the pixels in the target local region.

7 . The road obstacle detection device according to claim 1 , wherein
the region segmentation unit segments the image with a plurality of particle sizes, the likelihood calculation unit calculates the probability of the presence of the road obstacle in the target local region for each of segmentation results with the respective particle sizes, and the road obstacle detection device further comprises an integration unit configured to integrate the respective probabilities together to determine the probability of the presence of the road obstacle for each pixel.

8 . The road obstacle detection device according to claim 1 , further comprising a detection unit configured to subject threshold processing to the probability of the presence of the road obstacle, and detect a region in the image in which the road obstacle is present based on a result of the threshold processing.

9 . The road obstacle detection device according to claim 8 , further comprising a display configured to display an image in which a highlighted display to highlight the region detected by the detection unit is superimposed on the input image.

10 . The road obstacle detection device according to claim 1 , further comprising:
a semantic label estimation unit configured to estimate semantic labels of the plurality of local regions, wherein the likelihood calculation unit calculates a probability L i of presence of the road obstacle in a target local region S i based on an equation below

L
i

=

∑

j
=
1

N



{

n


(

S
j

)

·

d
appear



(

S
i

,

S
j

)

·

P
road



(

S
j

)

·

W


(

d
position



(

S
i

,

S
j

)

)

}

·

P
others



(

S
i

)

wherein
S i represents an i(=1 . . . ,N)-th local region,
n(S j ) represents a size of a local region S j ,
d appear (S i ,S j ) represents a degree of a visual difference between the local region S i and the local region S j ,
P road (S j ) represents a probability that the lo

Assignees

• Toyota Motor Co Ltd

Inventors

• Yamanaka Masao
• Kusumoto Yasuyuki

Classifications

• G06V20/58 Primary

  Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads · CPC title

• G08G1/16 Primary

  Anti-collision systems (road vehicle drive control systems for predicting or avoiding probable or impending collision otherwise than by control of a particular sub-unit B60W30/08) · CPC title

• G06F18/2415

  based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate · CPC title

• G06V10/768

  using context analysis, e.g. recognition aided by known co-occurring patterns · CPC title

• G06V10/255

  Detecting or recognising potential candidate objects based on visual cues, e.g. shapes · CPC title