剛体の姿勢推定

本資料は2020年10月10日に社内共有資料として展開していたものをWEBページ向けにリニューアルした内容になります。

■Object Pose Estimation

• Correspondence-based method : Find Correspondences between 2D points and 3D points (PnP, ICP)
• Template-based method: Extract the gradient information for matching, and refined(ICP)
• Voting-based method: Every local predicts a result, and refined using RANSAC
• Regression-based method: Represent pose suitable for CNN

■Summary of 6D pose estimation methods

Arxiv:1905.06658

■Correspondence-based method

With rich textures to get image features

Arxiv:1905.06658

■PnP solver

Estimate extrinsic camera parameter using known 3D points in a world coordinate

There are many many solvers (e.g. P3P, P5P.., using RANSAC or Eigen value decomposition)

Arxiv:1905.06658

■ICP algorithm

Iterative matching for 2 point clouds

1. Correspondence nearest neighbor points
2. Translate a point cloud set to match the other sets
3. iteration above

■ICP

Estimate extrinsic camera parameter using known 3D points in a world coordinate There are many many solvers (e.g. P3P, P5P.., using RANSAC or Eigen value decomposition)

Arxiv:1905.06658

■SegICP　SegNet

SegICP

　　Segmentation and depth image-based partial Registration to obtain object’s pose

SegNet
　　Multi-hypothesis object pose: Point-to-point ICP

　　1cm pose error and < 5°error

■Template-based method

Using gradient information in RGB-D or RGB image Without rich texture

ICP

Arxiv:1905.06658

■Voting-based method

Objects with occlusions Dense class labeling

Arxiv:1905.06658

■Regression-based method

PoseCNN

• Localizing the center of objects
• Predicting its distance from camera
• Regressing rotation to a quarternion

PoseCNN architecture

• Feature extraction
• Embedding
• Classification and regression

Two loss function for quaternion

• PoseLoss
• ShapeMatch-Loss

DeepIM
Iterative refinement using DNN
Enable to combine other estimator

DeepIM
Iterative refinement using DNN

Flownet: estimate optical flow

CullNet
Single view image-based object pose estimation
Culling false positives Calibrate the confidence score

■Correspondence-based and Regression-based method

CullNet Architecture

Using Yolov3 for keypoints

Cullnet output confidence of objects pose by PnP.

■Voting-based and Regression-based method

DenseFusion
Estimating 6D pose of known objects

■Comparison

ADD (Average Distance of Model Point)
ADD-S (for symmetric objects

Error smaller than threshold(< 2cm)

From the tables, we can see that DenseFusion achieved the highest accuracy comparing with other regression-based methods. novel local feature fusion scheme using both RGB and depth images

■ダウンロード

剛体の姿勢推定.pdf