Camera Pose Estimation and Relocalization Algorithm Based on Ground Texture

In recent years, mobile robots have become progressively more prevalent in daily life. Throughout their usage, the accuracy and robustness of the system can be greatly affected by changes in the peripheral environment. However, the system has not fully utilized the available ground information. This article presents a solution that utilizes ground texture information to achieve accurate mapping and localization. The proposed solution encompasses the development of a comprehensive system that incorporates plane map construction, pose estimation, and automatic relocalization functions. The system effectively optimizes pose and mapping effects by incorporating artificial loop frames. Furthermore, it successfully achieves the automatic relocalization with a mismatch rate of less than 1% through the use of feature matching based on a prior map. This effectively resolves the issue of losing track of the target. This article conducts experiments on various ground texture features to thoroughly validate the accuracy and robustness of the system.

I. Introduction

Artificial intelligence technology has led to the widespread use of autonomous mobile robots in various fields, such as manufacturing, warehouse logistics, and home services [1], [2]. Positioning technology is a crucial module for mobile robots [3]. Prior to the emergence of laser and vision solutions, magnetic navigation was predominantly utilized in traditional positioning systems. This solution is susceptible to magnetic field interference during use, requiring frequent maintenance and not being suitable for large-area scenes. Therefore, it is gradually being replaced [4], [5]. Next, the laser positioning solution was introduced. The laser positioning solution offers superior accuracy, but it comes with a higher cost of use [6], [7]. Thus, the implementation of this solution is subject to certain limitations.