# US 2013/0156295 A1 · Method of filtering a disparity mesh obtained from pixel images

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

- Number: US 2013/0156295 A1
- Document ID: US20130156295A1
- Status: Published application
- Kind: A1
- Application number: 13/442,328
- Filed: 2012-04-09
- Published: 2013-06-20
- Earliest claimed: 2011-12-14
- Assignees: Thomson Licensing
- Inventors: Paul Kerbiriou; Guillaume Boisson; Thierry Viellard
- Web page: https://patentgrep.com/patent/US20130156295A1

## CPC

- G06V 10/30: Noise filtering
- G06T 5/70: Denoising; Smoothing
- G06V 20/64: Three-dimensional [3D] objects
- G06T 2207/10028: Range image; Depth image; 3D point clouds
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

A method of filtering a disparity mesh from pixel images according to the invention, where the disparity mesh comprises a plurality of points, where each point is associated with values of two planar coordinates (X, Y) and a disparity value (D) and where the values are quantization pitches, comprises the step: filtering planes by filtering 2D-lines in 2D-spaces (X-D, Y-D) of the planar coordinates (X,Y) and the disparity (D).

## Claims

### Claim 1

Method of filtering a disparity mesh obtained from pixel images,
the disparity mesh comprising a plurality of points, each point being associated with values of two planar coordinates and a disparity value, the values being quantization pitches, wherein the method comprises the step:
filtering planes by filtering 2D-lines in 2D-spaces of the planar coordinates and the disparity.

### Claim 2 (depends on 1)

Method according to claim 1, wherein
the step filtering planes comprises:
for 2D-lines in the 2D-spaces,
detecting a 2D-line with at least two neighbour points,
iterative, determining line points of the detected 2D-line and adapting the 2D-line, and
modifying the line points to 2D-line locations of the 2D-line.

### Claim 3 (depends on 2)

Method according to claim 2, wherein a 2D-line with at least two neighbour points is detected if the 2D-line with the at least two neighbour points comply a line equation with the parameters slope (a) and constant (b) in the 2D-space:
D=a*Y+b,X=const. or
D=a*X+b,Y=const.

### Claim 4 (depends on 2)

Method according to claim 2, wherein
an examined point is determined as a line point of the 2D-line if an absolute difference of the disparity between the examined point and a pre-determined line point is less than or equal to one quantization pitch and if its distance from the 2D-line is lower or equal to a threshold.

### Claim 5 (depends on 4)

Method according to claim 4, wherein
an examined point is determined as a line point of the 2D-line if its distance from the 2D-line, which is adapted including the examined point, is lower or equal to a threshold.

### Claim 6 (depends on 2)

Method according to claim 2, wherein
the 2D-line with line points is defined with a normalized direction vector which is the normalized Eigen vector of the covariance matrix of the line points with the greatest Eigen value and a gravity centre of the line points.

### Claim 7 (depends on 2)

Method according to claim 2, wherein
a line point is modified to a 2D-line location of the 2D-line by projecting the line point onto the 2D-line.

### Claim 8 (depends on 7)

Method according to claim 7, wherein
the line point is projected onto the 2D-line by computing the scalar product of the vector from the gravity centre of the line points to the line point with the normalized direction vector of the 2D-line and adding to the gravity centre the normalized direction vector multiplied by the scalar product.

### Claim 9 (depends on 1)

Method according to claim 1, wherein
the method comprises the step of:
filtering contours in the 3D-space before filtering planes.

### Claim 10 (depends on 9)

Method according to claim 9, wherein
the step filtering contours in the 3D-space comprises:
detecting contour points of contours,
iterative, determining contour points to segment points of a 3D-line segment of a 3D-line and adapting the 3D-line, and,
modifying the segment points to 3D-line points of the 3D-line.

### Claim 11 (depends on 10)

Method according to claim 10, wherein
an examined point is detected as a contour point if an absolute difference of the disparity between the examined point and a surrounding point is greater than one quantization pitch.

### Claim 12 (depends on 10)

Method according to claim 10, wherein
a contour point is determined as a segment point of the 3D-line segment of the 3D-line if an absolute difference of the disparity between the contour point and a pre-determined segment point is less than or equal to one quantization pitch and if its distance from the 3D-line is lower or equal to a threshold.

### Claim 13 (depends on 10)

Method according to claim 10, wherein
the 3D-line is defined with a normalized direction vector which is the normalized Eigen vector of the covariance matrix of the segment points with the greatest Eigen value and a gravity centre of the segment points.

### Claim 14 (depends on 10)

Method according to claim 10, wherein
a segment point is modified to a 3D-line point of the 3D-line by projecting the segment point onto the 3D-line.

### Claim 15 (depends on 14)

Method according to claim 14, wherein
a segment point is projected onto the 3D-line by computing the scalar product between the vector from the gravity centre to the segment point and the normalized direction vector of the 3D-line and adding to the gravity centre the normalized direction vector multiplied by the scalar product.
