A Human Behavior Analysis based on Motion Trajectory Measurement for Guiding Services

Hajime Asama, A.k. Morimoto, Makoto Nishimura, Tomokazu MishimaThe Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec) 2004 · DOI

Abstract

In the evolving field of Service Engineering, the ability to provide timely and relevant assistance in public spaces is paramount. This paper presents a non-intrusive system developed by researchers at the University of Tokyo and Saitama University designed to distinguish between "insiders" (individuals familiar with their surroundings) and "visitors" (those potentially requiring guidance). By analyzing walking trajectories via a specialized vision chip and a novel "motion index" ($v \cos \theta$), the system achieves over 90% reliability in identifying individuals in need of assistance. This approach offers a privacy-preserving and efficient alternative to traditional interactive kiosks or manual observation.

Introduction

A central challenge in automated service delivery is determining the optimal moment to offer assistance. In large-scale environments such as airports, hospitals, or corporate campuses, providing guidance to every occupant is both computationally inefficient and socially intrusive. To address this, the authors propose a framework for intent recognition based on observable physical behavior.

The core hypothesis is that a person's familiarity with an environment is encoded in their movement patterns. A staff member (an "insider") typically moves with purpose and a linear trajectory, while a lost visitor often exhibits slower speeds and frequent changes in direction as they scan for landmarks or signage. By quantifying these differences, the researchers aim to create an ambient intelligence system capable of triggering guidance services only when a genuine need is detected.

Methodology

Hardware and Image Processing

The system utilizes a specialized "retina chip"—a vision sensor that performs initial image processing on-chip. This hardware choice provides two distinct advantages: high processing speeds and enhanced privacy, as the system focuses on coordinate data rather than high-resolution imagery.

The processing pipeline involves:

  1. Trajectory Measurement: Tracking the subject's movement through the environment.
  2. Background Subtraction: Isolating the moving subject from the static environment.
  3. Coordinate Transformation: Mapping 2D image pixels to real-world floor coordinates to ensure accurate spatial analysis.

The Motion Index

The researchers developed a custom metric to differentiate between purposeful and exploratory movement. This "motion index" is defined by the relationship between velocity ($v$) and the change in direction ($\theta$). The formula used is:

$$\text{Motion Index} = v \cos \theta$$

This metric effectively penalizes both low speeds and erratic directional shifts. A high index value indicates a person moving quickly in a straight line, whereas a low or fluctuating value suggests a person who is wandering or hesitant.

Statistical Classification

To classify individuals, the authors modeled the distribution of the motion index for both insiders and visitors. By approximating these distributions as probability density functions (specifically normal distributions), the system can establish a statistical threshold. If an individual's motion index falls within the visitor distribution with high probability, the system identifies them as a candidate for guidance services.

Experimental Results

Measurement Accuracy

Validation of the tracking system demonstrated high fidelity, with a measurement error of less than 350mm at a distance of 6 meters. This level of precision is sufficient to capture the subtle nuances in walking trajectories required for the motion index calculation.

Classification Reliability

In preliminary experiments conducted in an indoor hallway environment, the system demonstrated remarkable accuracy. By comparing the movement of participants familiar with the layout against those who were not, the system was able to distinguish between the two groups with over 90% reliability. The data confirmed that the motion index for both groups follows a near-normal distribution, validating the probabilistic approach to classification.

Discussion and Future Work

Movement as Communication

The study highlights that intent is often a byproduct of physical movement. A "wandering" trajectory is not merely noise in the data; it is a signal of high cognitive load or a lack of spatial information. By treating movement as a form of non-verbal communication, service systems can become more proactive and less reliant on user-initiated queries.

Limitations

While the results are promising, the current implementation is limited to single-subject tracking in controlled environments. Future research will focus on:

  • Multi-Subject Tracking: Identifying and tracking individuals within crowded scenes.
  • Wide-Area Monitoring: Coordinating multiple cameras to maintain trajectory data across larger, more complex architectural spaces.
  • Expanded Applications: Applying this trajectory analysis to security contexts (e.g., detecting loitering) or retail environments to better understand consumer behavior.

Conclusion

By combining specialized vision hardware with an elegant mathematical model of human movement, the authors have demonstrated a robust method for intent recognition. This system paves the way for smarter, more responsive environments that can offer help exactly when it is needed, without compromising user privacy or comfort.

Summary updated

BibTeX
@inproceedings{asama2004human,
  author = {Hajime Asama and A.K. Morimoto and Makoto Nishimura and Tomokazu Mishima},
  title = {{A Human Behavior Analysis based on Motion Trajectory Measurement for Guiding Services}},
  booktitle = {The Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec)},
  year = {2004},
  doi = {10.1299/jsmermd.2004.203\_5},
}
Figure