A Projective Geometric View for 6D Pose Estimation in mmWave MIMO Systems

TL;DR

This paper introduces a novel approach to 6D pose estimation in mmWave MIMO systems by connecting AoA/AoD measurements with computer vision's perspective projection, enabling near-optimal localization and SLAM even without LoS.

Contribution

It establishes a new link between mmWave localization and computer vision models, allowing the application of advanced pose estimation techniques to mmWave systems.

Findings

Estimators operate close to theoretical bounds.

Effective 6D SLAM without LoS or LoS/NLoS info.

Connection enables new localization methods.

Abstract

Millimeter-wave (mmWave) systems in the 30--300 GHz bands are among the fundamental enabling technologies of 5G and beyond 5G, providing large bandwidths, not only for high data rate communication, but also for precise positioning services, in support of high accuracy demanding applications such as vehicle positioning. With the possibility to introduce relatively large arrays on user devices with a small footprint, the ability to determine the user orientation becomes unlocked. The estimation of the full user pose (joint 3D position and 3D orientation) is referred to as 6D localization. Conventionally, the problem of 6D localization using antenna arrays has been considered difficult and was solved through a combination of heuristics and optimization. In this paper, we reveal a close connection between the AoA and AoD and the well-studied perspective projection model from computer vision. This connection allows us to solve the 6D localization problem, by adapting state-of-the-art methods from computer vision. More specifically, two problems, namely 6D pose estimation from AoA from multiple single-antenna base stations and 6D SLAM based on single-BS mmWave communication, are first modeled with the perspective projection model, and then solved. Numerical simulations show that the proposed estimators operate close to the theoretical performance bounds. Moreover, the proposed SLAM method is effective even in the absence of the LoS path, or knowledge of the LoS/NLoS condition.