# US 11,443,526 B2 · Road obstacle detection device, road obstacle detection method, and recording medium storing a road obstacle detection program

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 11,443,526 B2
- Document ID: US11443526B2
- Status: Granted
- Kind: B2
- Application number: 17/002,995
- Filed: 2020-08-26
- Published: 2021-04-15
- Granted: 2022-09-13
- Earliest claimed: 2019-10-10
- Assignees: Toyota Jidosha Kabshiki Kaisha
- Applicants: Toyota Jidosha Kabushiki Kaisha
- Inventors: Toshiaki Ohgushi; Kenji Horiguchi; Masao Yamanaka
- Web page: https://patentgrep.com/patent/US11443526B2

## CPC

- G01V 8/00: Prospecting or detecting by optical means
- G01V 8/10: Detecting, e.g. by using light barriers (by reflection from the object G01S17/00)
- G06T 7/11: Region-based segmentation
- G06V 10/25: Determination of region of interest [ROI] or a volume of interest [VOI]
- G06V 10/82: using neural networks
- G06V 20/58: Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V 30/274: {Syntactic or semantic context, e.g. balancing}
- G06V 30/2528: {Combination of methods, e.g. classifiers, working on the same input data}
- G01V: GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS (means for indicating the location of accidentally buried, e.g. snow-buried, persons A63B29/02)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A road obstacle detection device which uses a pre-learned first identifier to associate a semantic label with each pixel of an image, uses a pre-learned second identifier to estimate a statistical distribution of a semantic label of a predetermined region of interest of the image from a statistical distribution of a semantic label of a peripheral region that surrounds the region of interest, and uses the statistical distribution of the semantic label associated with the region of interest and the statistical distribution of the semantic label estimated for the region of interest to estimate a likelihood that an object is a road obstacle.

## Claims

### Claim 1

A road obstacle detection device, comprising:
a memory; and
a processor coupled to the memory,
wherein the processor is configured to:
use a pre-learned first identifier to associate a semantic label with each pixel of an image;
use a pre-learned second identifier to estimate a statistical distribution of a semantic label of a region of interest of the image from a statistical distribution of a semantic label of a peripheral region that surrounds the region of interest, the region of interest being predefined; and
use a statistical distribution of the semantic label associated with the region of interest and the statistical distribution of the semantic label estimated for the region of interest to estimate a likelihood that an object is a road obstacle, wherein
the likelihood that the object is a road obstacle is estimated based on a square of a difference between the statistical distribution of the semantic label associated with the region of interest and the statistical distribution of the semantic label estimated for the region of interest, an absolute value of the difference, an inner product, and a probability distribution distance.

### Claim 2 (depends on 1)

The road obstacle detection device according to claim 1, wherein the processor is configured to:
divide the image into a plurality of localized regions located in only one of a foreground or a background;
select the region of interest and the peripheral region from the plurality of localized regions; and
estimate the likelihood that the object is a road obstacle.

### Claim 3 (depends on 1)

The road obstacle detection device according to claim 1, wherein the region of interest is a rectangular region that encompasses a region likely to be an object, which is acquired by extracting a region of a solid mass from an image.

### Claim 4

A road obstacle detection method, the method comprising, by a processor:
using a pre-learned first identifier to associate a semantic label with each pixel of an image, and using a pre-learned second identifier to estimate a statistical distribution of a semantic label of a predetermined region of interest of the image from a statistical distribution of a semantic label of a peripheral region that surrounds the region of interest, wherein the region of interest is predetermined; and
using a statistical distribution of the semantic label associated with the region of interest and the statistical distribution of the semantic label estimated for the region of interest to estimate a likelihood that an object is a road obstacle, wherein
the likelihood that the object is a road obstacle is estimated based on a square of a difference between the statistical distribution of the semantic label associated with the region of interest and the statistical distribution of the semantic label estimated for the region of interest, an absolute value of the difference, an inner product, and a probability distribution distance.

### Claim 5 (depends on 4)

The road obstacle detection method according to claim 4, the method comprising, by a processor:
dividing the image into a plurality of localized regions located in only one of a foreground or a background;
selecting the region of interest and the peripheral region from the plurality of localized regions; and
estimating the likelihood that the object is a road obstacle.

### Claim 6 (depends on 4)

The road obstacle detection method according to claim 4, wherein the region of interest is a rectangular region that encompasses a region likely to be an object, which is acquired by extracting a region of a solid mass from an image.

### Claim 7

A non-transitory storage medium that stores a program that causes a processor to execute road obstacle detection processing, the processing including:
using a pre-learned first identifier to associate a semantic label with each pixel of an image, and using a pre-learned second identifier to estimate a statistical distribution of a semantic label of a region of interest of the image from a statistical distribution of a semantic label of a peripheral region that surrounds the region of interest, wherein the region of interest is predetermined; and
using a statistical distribution of the semantic label associated with the region of interest and the statistical distribution of the semantic label estimated for the region of interest to estimate a likelihood that an object is a road obstacle, wherein
the likelihood that the object is a road obstacle is estimated based on a square of a difference between the statistical distribution of the semantic label associated with the region of interest and the statistical distribution of the semantic label estimated for the region of interest, an absolute value of the difference, an inner product, and a probability distribution distance.

### Claim 8 (depends on 7)

The non-transitory storage medium according to claim 7, wherein processing further comprises:
dividing the image into a plurality of localized regions located in only one of a foreground or a background;
selecting the region of interest and the peripheral region from the plurality of localized regions; and
estimating the likelihood that the object is a road obstacle.

### Claim 9 (depends on 7)

The non-transitory storage medium according to claim 7, wherein the region of interest is a rectangular region that encompasses a region likely to be an object, which is acquired by extracting a region of a solid mass from an image.