Robust improvement solution to perspective-n-point problem

Feng Youyang,Wang Qing,Yan Chao,Yang Yuan

doi:10.1177/1729881419885700

Feng Youyang, Wang Qing + Show 2 more

Open Access

https://doi.org/10.1177/1729881419885700

Copy DOI

Abstract

Perspective-n-point is a classical computer vision problem that uses three-dimensional points and image pixels to estimate camera pose. The visual robot often loses its position when the camera moves too fast or the environment changes. Perspective-n-point is used to relocate robot position, but the distribution of three-dimensional points in the world frame and different choices of coordinates affect the perspective-n-point performance and make perspective-n-point results less robust and inaccurate. In this study, we review the previous perspective-n-point algorithms and provide their disadvantages when facing three-dimensional points with large variances. According to the drawbacks of previous perspective-n-point methods, we propose a normalization method inspired by the homogeneous matrix calculation process to increase perspective-n-point algorithm accuracy and robustness. The experimental results demonstrate that the proposed perspective-n-point method is robust to different choices of coordinates and is thus better than other state-of-art perspective-n-point methods. Considering that the true camera pose is difficult to obtain, the former perspective-n-point solution validation experiment is mostly based on simulated image data. In this study, we design a new experiment based on total station and chessboard to verify the robustness and accuracy of the perspective-n-point algorithm.

Highlights

The PnP problem focuses on reconstructing camera absolute pose using known image pixels and three-dimensional (3-D) points in world space, which is a widely studied problem in computer vision,9 such as in augmented reality,10 simultaneous localization and mapping (SLAM) pose recovery in loss of frames,11 and vision localization in smart cities
The previous PnP experiments based on real images only established correspondences by matching image feature points and estimating camera pose using the PnP algorithm
The additional computation time compared with other PnP algorithms is the time cost associated with principal component analysis (PCA)

Summary

Introduction

There are three main geometry problems in multiple computer vision: the homography problem, iterative closest point (ICP) problem, and perspective-n-point (PnP) problem. The PnP problem focuses on reconstructing camera absolute pose using known image pixels and three-dimensional (3-D) points in world space, which is a widely studied problem in computer vision, such as in augmented reality, simultaneous localization and mapping (SLAM) pose recovery in loss of frames, and vision localization in smart cities. Different selections of known 3-D points does not affect PnP algorithm results very much, and the absolute pose in the world should be close to the true value. Previous PnP algorithms show high accuracy using simulated data, but the performance is very poor in our designed experiment. In the “Normalization of PnP method” section, we describe the details of the proposed PnP algorithm and analyze the reason it performs better than state-of-art PnP methods. In the “Results” section, which demonstrate that previous PnP algorithms cannot perform very well if 3-D points are distributed unequally. Experimental results presented in the “Results” section show that our proposed method is better than the state-of-art PnP algorithm, especially in terms of robustness in a real scene.

Related work

ÃT ð7Þ

Results

Objective

Conclusions