Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DoorINet: Door Heading Prediction through Inertial Deep Learning

Aleksei Zakharchenko, Sharon Farber, Itzik Klein

TL;DR

DoorINet tackles indoor door-heading estimation without magnetometers by learning end-to-end mappings from door-mounted IMU data. It introduces two networks, AG-DoorINet (accelerometer+gyroscope) and G-DoorINet (gyroscope-only), and validates them on a unique 391-minute dataset collected from multiple doors, outperforming model-based and data-driven baselines. The study demonstrates strong generalization across sensor variations and long-duration operation, achieving sub-degree RMSE in several test cases. This magnetometer-free approach enables reliable heading estimation for door-related IoT applications in smart homes and offices, with reproducibility supported by publicly available code and data.

Abstract

Inertial sensors are widely used in a variety of applications. A common task is orientation estimation. To tackle such a task, attitude and heading reference system algorithms are applied. Relying on the gyroscope readings, the accelerometer measurements are used to update the attitude angles, and magnetometer measurements are utilized to update the heading angle. In indoor environments, magnetometers suffer from interference that degrades their performance resulting in poor heading angle estimation. Therefore, applications that estimate the heading angle of moving objects, such as walking pedestrians, closets, and refrigerators, are prone to error. To circumvent such situations, we propose DoorINet, an end-to-end deep-learning framework to calculate the heading angle from door-mounted, low-cost inertial sensors without using magnetometers. To evaluate our approach, we record a unique dataset containing 391 minutes of accelerometer and gyroscope measurements and corresponding ground-truth heading angle. We show that our proposed approach outperforms commonly used, model based approaches and data-driven methods.

DoorINet: Door Heading Prediction through Inertial Deep Learning

TL;DR

DoorINet tackles indoor door-heading estimation without magnetometers by learning end-to-end mappings from door-mounted IMU data. It introduces two networks, AG-DoorINet (accelerometer+gyroscope) and G-DoorINet (gyroscope-only), and validates them on a unique 391-minute dataset collected from multiple doors, outperforming model-based and data-driven baselines. The study demonstrates strong generalization across sensor variations and long-duration operation, achieving sub-degree RMSE in several test cases. This magnetometer-free approach enables reliable heading estimation for door-related IoT applications in smart homes and offices, with reproducibility supported by publicly available code and data.

Abstract

Inertial sensors are widely used in a variety of applications. A common task is orientation estimation. To tackle such a task, attitude and heading reference system algorithms are applied. Relying on the gyroscope readings, the accelerometer measurements are used to update the attitude angles, and magnetometer measurements are utilized to update the heading angle. In indoor environments, magnetometers suffer from interference that degrades their performance resulting in poor heading angle estimation. Therefore, applications that estimate the heading angle of moving objects, such as walking pedestrians, closets, and refrigerators, are prone to error. To circumvent such situations, we propose DoorINet, an end-to-end deep-learning framework to calculate the heading angle from door-mounted, low-cost inertial sensors without using magnetometers. To evaluate our approach, we record a unique dataset containing 391 minutes of accelerometer and gyroscope measurements and corresponding ground-truth heading angle. We show that our proposed approach outperforms commonly used, model based approaches and data-driven methods.
Paper Structure (27 sections, 14 equations, 14 figures, 4 tables)

This paper contains 27 sections, 14 equations, 14 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Heading Estimation Approaches
  3. Model-based approaches
  4. Direct Integration
  5. Madgwick Filter
  6. Learning approaches
  7. Deep Attitude Estimator (DAE) EranDAE
  8. Quaternion Model A ASGHARPOORGOLROUDBARI2023113105
  9. LGC-Net lgcnet
  10. Proposed Learning Approach for Heading Estimation
  11. Gyroscope-Only DoorINet Architecture (G-DoorINet)
  12. Accelerometer and Gyroscope DoorINet Architecture (AG-DoorINet)
  13. Dataset generation and processing
  14. Train and validation datasets
  15. Session 1
  16. ...and 12 more sections

Figures (14)

  • Figure 1: Our proposed data-driven approach.
  • Figure 2: G-DoorINet network architecture using only gyroscope measurements.
  • Figure 3: AG-DoorINet network architecture using accelerometer and gyroscope measurements.
  • Figure 4: Experimental setups used for dataset generation: DOT IMUs generated raw IMU readings and Memsense IMU generated the ground truth.
  • Figure 5: An example of data recorded in Session 1.
  • ...and 9 more figures