FTIN: Frequency-Time Integration Network for Inertial Odometry

TL;DR

The paper addresses inertial odometry under high-rate IMU signals, where redundancy induces an information bottleneck in trajectory estimation. It introduces FTIN, a dual-domain fusion framework that combines Frequency-Domain Learning (FDL) and Time-Domain Learning (TDL) to extract global context and long-range dependencies from IMU data, with FD L leveraging FFT along the channel axis to produce energy-efficient frequency features and an iFFT to recover time-domain signals, and TD L using a scalar LSTM to model temporal structure. The architecture uses a ResNet-1D backbone, followed by an MLP head to predict velocity, and demonstrates improvements across six public datasets, including large reductions in ATE, RTE, and PDE compared to baselines. This work provides a new perspective on IO by exploiting energy compaction in the frequency domain to complement time-domain processing, enabling more robust localization in practice, though cross-platform IMU variability remains to be explored. Overall, FTIN establishes that cross-domain fusion can alleviate information bottlenecks in high-rate IO and yield practical gains for real-world navigation systems.

Abstract

Inertial odometry (IO) leverages inertial measurement unit (IMU) signals for cost-effective localization. However, high IMU sampling rates introduce substantial redundancy that impedes IO's ability to attend to salient components, thereby creating an information bottleneck. To address this challenge, we propose a cross-domain IO framework that fuses information from the frequency and time domains. Specifically, we exploit the global context and energy-compaction properties of frequency-domain representations to capture holistic motion patterns and alleviate the bottleneck. To the best of our knowledge, this is among the first attempts to incorporate frequency-domain feature processing into IO. Experimental results on multiple public datasets demonstrate the effectiveness of the proposed frequency--time-domain fusion strategy.

Figures (4)

• Figure 1: Comparison on the RoNIN dataset. Algorithms closer to the lower-left corner indicate lower errors.
• Figure 2: Network Architecture Diagram of FTIN.
• Figure 3: PDE comparison between FTIN and three baseline models from RoNIN on the RoNIN dataset.
• Figure 4: Visual comparison of the FTIN and RoNIN ResNet on various datasets.