Fast Optimal Joint Tracking-Registration for Multi-Sensor Systems

TL;DR

This paper introduces a fast, stable, and asymptotically optimal algorithm for joint registration and tracking in multi-sensor systems, significantly improving accuracy and efficiency in sensor fusion for vehicular applications.

Contribution

The paper presents a novel FMAP algorithm that combines recursive optimization, orthogonal factorization, and Givens rotation for efficient joint registration-tracking.

Findings

Algorithm achieves O(n) complexity for multiple targets.

Demonstrates asymptotic optimality via Cramér-Rao bounds.

Shows improved accuracy and speed in simulations and experiments.

Abstract

Sensor fusion of multiple sources plays an important role in vehicular systems to achieve refined target position and velocity estimates. In this article, we address the general registration problem, which is a key module for a fusion system to accurately correct systematic errors of sensors. A fast maximum a posteriori (FMAP) algorithm for joint registration-tracking (JRT) is presented. The algorithm uses a recursive two-step optimization that involves orthogonal factorization to ensure numerically stability. Statistical efficiency analysis based on Cram\`{e}r-Rao lower bound theory is presented to show asymptotical optimality of FMAP. Also, Givens rotation is used to derive a fast implementation with complexity O(n) with $n$ the number of tracked targets. Simulations and experiments are presented to demonstrate the promise and effectiveness of FMAP.