Exploration and Implementation of a Pre-Impact Fall Recognition Method Based on an Inertial Body Sensor Network

Guoru Zhao, Zhanyong Mei, Ding Liang, Kamen Ivanov, Yanwei Guo, Yongfeng Wang, Lei Wang2012 · DOI

TL;DR

This paper presents a novel pre-impact fall recognition method using an inertial body sensor network to predict falls before they happen, aiming to reduce fall-related injuries. By analyzing kinematic data from sensors placed on the torso, the study identifies the chest as the optimal location for early fall detection and establishes specific acceleration thresholds for both pre-impact alarms and post-fall detection.

What Problem Does This Paper Address?

Falls are a significant cause of injury and death, particularly among the elderly, leading to substantial health and economic burdens. Early detection and alarm systems that can predict an impending fall before it occurs are crucial for enabling self-protective measures or interventions to mitigate injury severity. Existing methods often struggle with achieving a sufficient "pre-impact lead time" – the window of opportunity between detection and impact – and accurately distinguishing falls from normal daily activities. This paper tackles the challenge of developing a reliable and practical pre-impact fall recognition system using wearable inertial sensors, focusing on identifying optimal sensor placement and detection parameters to maximize this critical lead time.

Methods

The researchers developed a pre-impact fall recognition/alarm method based on an inertial body sensor network. This approach utilizes multiple sensor nodes placed on different body segments to capture kinematic data. Key methodological aspects include:

  • Sensor Network: Nine MTx inertial sensor modules (measuring 3D acceleration, 3D angular rate, and 3D Earth-magnetic field) were used.
  • Sensor Placement Optimization: The study systematically analyzed data from various body segments (chest, fore-waist, left waist, left thigh, left shank, left foot, right thigh, right shank, and right foot) to determine the optimal location for fall detection.
  • Feature Extraction: Kinematic features, specifically acceleration and angular rate, were extracted from the sensor data.
  • Threshold Determination: Threshold levels for acceleration and angular rate were established for both early pre-impact alarm and post-fall detection.
  • Pre-Impact Lead Time and Postural Stability Analysis: The study defined and calculated the pre-impact lead time (T) and the critical angle of body postural stability (θ) to understand the dynamics of falls and the potential for intervention.
  • Comparative Analysis: The effectiveness of acceleration versus angular rate as input parameters for fall recognition was compared.

Datasets

The study utilized data collected from experiments involving eight healthy Asian adult subjects. These subjects performed a set of Activities of Daily Living (ADL) and Fall Activities (FA).

  • ADL: Stand-Sit-Stand (SSS), Walking, Stand-Sit-Lie (SSL).
  • FA: Right-sideway Fall (RF), Forward Fall (FF), Backward Fall (BF).

Each subject performed each activity three times, resulting in a total of 22 trials. This dataset was used for training and evaluating the proposed fall recognition method. No pre-existing public datasets were explicitly mentioned as being used for training or comparison.

Key Insights and Ideas

  • Chest is King (for early detection): The chest emerged as the optimal sensor placement for early pre-impact fall recognition and post-fall detection. This is a significant finding as it suggests a single sensor on the torso might be sufficient for effective fall monitoring, simplifying system design and user comfort.
  • Acceleration Over Angular Rate for Early Warning: Acceleration data proved to be a more effective input parameter than angular rate for early pre-impact fall recognition, offering a longer lead time. This highlights the importance of focusing on linear motion changes in predicting an impending fall.
  • Distinct Fall Dynamics: Different types of falls (forward, sideway, backward) exhibit distinct kinematic profiles, particularly in terms of pre-impact lead time and postural stability angles. This understanding is crucial for tailoring detection algorithms and intervention strategies.
  • Backward Falls are the Toughest: Backward falls present the most significant challenge for avoidance due to a combination of the shortest pre-impact lead time and the smallest postural stability angle, making self-protective mechanisms less effective.

Key Results

  • Optimal Sensor Placement: The chest was identified as the optimal sensor placement for fall recognition/alarm.
  • Thresholds for Detection:
    • Early pre-impact alarm: 7 m/s/s (acceleration).
  • Post-fall detection: 20 m/s/s (acceleration).
  • Pre-Impact Lead Times:
    • Forward Fall (FF): Longest average lead time of 329 ± 21 ms.
  • Right-sideway Fall (RF): 265 ± 35 ms.
  • Backward Fall (BF): Shortest average lead time of 257 ± 36 ms.
  • Critical Angles of Postural Stability:
    • Forward Fall (FF): 23.9 ± 3.3 degrees.
  • Right-sideway Fall (RF): 49.9 ± 4.1 degrees.
  • Backward Fall (BF): 9.9 ± 2.5 degrees.
  • Reliability: The chest placement achieved 100% reliability in recognizing fall activities during ADL when using a 20 m/s/s acceleration threshold.

Limitations and Future Work

  • Healthy Young Subjects: All experiments were conducted on healthy young adults under mat-bed cushioned impact conditions. The authors acknowledge that these results may not fully represent the real-world conditions of elderly falls, which can occur on harder surfaces and involve different physiological states.
  • Future Work: The researchers plan to conduct continuous 24-hour monitoring of individuals at high risk of falls in a natural environment, using a chest-mounted sensor, to acquire data on real elderly falls. This will allow for a more comprehensive evaluation and validation of their system.

Summary updated

BibTeX
@misc{zhao2012exploration,
  author = {Guoru Zhao and Zhanyong Mei and Ding Liang and Kamen Ivanov and Yanwei Guo and Yongfeng Wang and Lei Wang},
  title = {{Exploration and Implementation of a Pre-Impact Fall Recognition Method Based on an Inertial Body Sensor Network}},
  year = {2012},
  doi = {10.3390/s121115338},
}

Loading related content...

Figure