Robust pedestrian trajectory reconstruction from inertial sensor

Beaufils, Bertrand; Chazal, Frederic; Grelet, Marc; Michel, Bertrand · 2019 · Crossref

DOI: 10.1109/ipin.2019.8911819

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of robust pedestrian trajectory reconstruction using inertial measurement units (IMUs) worn on the ankle, specifically in environments where Global Navigation Satellite Systems (GNSS) fail, such as indoors or tunnels. The primary motivation is to overcome the limitations of existing stride detection methods, which often rely on fixed thresholds or sliding windows that struggle to identify atypical movements—such as small steps, side steps, backward walking, or stairs—and are prone to false positives during non-locomotive activities like sitting or bicycling. The authors propose a novel approach combining a robust sensor alignment technique with a machine learning-based stride classifier to improve accuracy in daily life contexts. The methodology begins with computing a terrestrial reference frame to align inertial data, correcting for gravity and sensor drift. The authors introduce a technique that estimates gravity by averaging acceleration over a 15-second window in an inertial frame, allowing for the removal of gravity from the acceleration data. This enables the computation of a "pseudo-speed" by integrating acceleration forward and backward from detected periods of inactivity (stance phases). These pseudo-speed intervals serve as candidate strides. To distinguish true strides from false detections, the authors employ a supervised machine learning classifier using the Gradient Boosting Tree algorithm. Features are engineered from the inertial data and pseudo-speed, including metrics derived from the assumption that ankle speed at the start and end of a stride corresponds to pure rotation around the foot's contact point. The algorithm was evaluated on a dataset comprising over 7,800 strides from seven subjects performing various activities, including walking, running, and climbing stairs. Validation was conducted using motion capture (MOCAP) systems to analyze stride length accuracy. Additionally, the system was tested in a challenging real-world scenario involving a 5.5-hour recording of an office worker. The results demonstrate that the proposed method successfully detects both normal and atypical strides, whereas traditional threshold-based methods failed to detect strides in narrow corridors or during stair climbing. The machine learning component effectively filtered out false positives generated by non-walking movements, such as ankle motion while seated. The significance of this work lies in its ability to provide robust trajectory reconstruction in complex, uncontrolled environments where standard inertial navigation systems degrade. By integrating a gravity-compensated reference frame with a data-driven classifier, the approach achieves higher reliability for dead reckoning in indoor positioning applications. This enables accurate localization in scenarios involving diverse gait patterns and mixed activities, addressing a critical gap in wearable navigation technology for daily use.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• pedestrian behavior perception