Method for three-dimensional human pose estimation

Patent metadata

Field	Value
Publication number	US-11200685-B2
Application number	US-201916724608-A
Country	US
Kind code	B2
Filing date	Dec 23, 2019
Priority date	Mar 18, 2019
Publication date	Dec 14, 2021
Grant date	Dec 14, 2021

Abstract

The invention discloses a method for three-dimensional human pose estimation, which can realize the real-time and high-precision 3D human pose estimation without high configuration hardware support and precise human body model. In this method for three-dimensional human pose estimation, including the following steps: (1) establishing a three-dimensional human body model matching the object, which is a cloud point human body model of visible spherical distribution constraint. (2) Matching and optimizing between human body model for human body pose tracking and depth point cloud. (3) Recovering for pose tracking error based on dynamic database retrieval.

First claim

What s claimed is:

1. A method for three-dimensional human pose estimation, comprising the following steps:
(1) establishing a three-dimensional human body model matching an object, wherein the three-dimensional human body model is a cloud point human body model of visible spherical distribution constraint;
(2) matching and optimizing between the cloud point human body model of visible spherical distribution constraint and a depth point cloud for human body pose tracking; and
(3) recovering a visible spherical distribution constraint point cloud manikin for pose tracking error based on dynamic database retrieval;
wherein in step (1):
representing a human body surface with 57 spheres, wherein each of the 57 spheres is characterized by a respective radius and a respective center, and the respective radius and the respective center are initialized empirically;
corresponding all of the 57 spheres to 11 body components to define a sphere set S to be a collection of 11 component sphere set models, wherein each of the 11 component sphere set models represents a body component and is defined by formula (1):

S
=

⋃
11

k
=
1

⁢

S
k

⁢

⁢

S
k

=

{

g
i
k

}

i
=
1

N
k

:=

{

[

c
i
k

,

r
i
k

]

}

i
=
1

N
k

(
1
)

wherein ∪ represents an operation of set union, g i k represents an ith sphere of a kth body component, c i k , r i k represent the respective center and the respective radius of the ith sphere in the kth body component, respectively, and N k represents a number of the spheres contained in the kth body component, with

∑

k
-
1

1
⁢
1

⁢

N
k

=

5
⁢

7
.

2. The method for the three-dimensional human pose estimation according to claim 1 , wherein in step (1), wrist and ankle movements are ignored.

3. The method for the three-dimensional human pose estimation according to the claim 2 , wherein in step (1), for all of the 57 spheres, constructing a directed tree,
wherein each node of the directed tree corresponds to a respective sphere, a root of the directed tree is g 1 1 , and each node has a unique parent node denoted by a black sphere, wherein a definition of the unique parent node is given by:
parent( S 1 )= g 1 1 ,parent( S 2 )= g 1 1 ,parent( S 3 )= g 3 2 ,parent( S 4 )= g 1 3 ,parent( S 5 )= g 1 2 ,parent( S 6 )= g 1 5 ,parent( S 7 )= g 2 2 ,parent( S 8 )= g 3 1 ,parent( S 9 )= g 1 8 ,parent( S 10 )= g 2 1 ,parent( S 11 )= g 1 10   (2)

wherein a motion of each body component is determined by a rotation motion R k in a local coordinate system with corresponding parent node as an origin plus a global translation vector t in a world coordinate system;
using a Fibonacci spherical algorithm to get spherical point cloud by dense sampling, wherein the cloud point human body model of visible spherical distribution constraint is as shown in formula (3):

=

⋃
11

k
=
1

⁢

:=

⋃
11

k
=
1

⁢

⋃

N
k

i
=
1

Assignees

• Univ Beijing Technology

Inventors

• Kong Dehui
• Wu Yongpeng
• Wang Shaofan
• Li Jinghua
• Wang Lichun

Classifications

• G06T17/20 Primary

  Finite element generation, e.g. wire-frame surface description, {tesselation} · CPC title

• Y02T10/40

  Engine management systems · CPC title

• G06T2207/10028

  Range image; Depth image; 3D point clouds · CPC title

• G06T2200/04

  involving 3D image data · CPC title

• G06T7/251 Primary

  involving models · CPC title