HulaMove: Using Commodity IMU for Waist Interaction

Title of the Paper

HulaMove: Using Commodity IMU for Waist Interaction

Paper Information

• Conference: CHI Conference on Human Factors in Computing Systems (CHI '21)
• Publication Date: May 8–13, 2021
• Research Area: Human-Computer Interaction (HCI)
• Keywords: Waist interaction, device-free sensing, inertial measurement unit (IMU), vision-free, all-weather interaction, virtual reality, augmented reality, daily task optimization, full-body interaction

Research Background and Problem

• Research Problem: Current human-computer interaction methods are gradually expanding from manual operations to inputs from different parts of the human body (e.g., face, feet). Most studies on waist motion focus on passively tracking users' daily activities, but the potential for active control interaction through the waist remains unexplored. As the largest joint in the human body, the waist can be easily controlled by users.
• Research Significance: Waist interaction offers a novel "vision-free, hands-free" input method, particularly useful in scenarios where users' eyes and hands are occupied (e.g., cooking or carrying heavy objects). It can also enhance immersion in virtual reality (VR).
• Related Work: Existing interaction studies based on other body parts (e.g., feet, hands, or ears) often rely on specialized hardware, and most IMU-based research imposes strict constraints on device placement. There is a lack of detailed interaction design and performance analysis for waist-based input in this field.

Solution

• Method Overview: A new interaction technique called HulaMove is proposed, leveraging the built-in IMU sensors of smartphones to recognize users' waist movements. The system enables real-time waist gesture recognition without additional hardware and supports interactions in both real-world and virtual reality scenarios.
• Innovations:
  • Interaction Design Space: User studies identified optimal gesture designs for waist interaction, suitable motion types, and the number of directions.
  • Lightweight Model and Versatility: A simple calibration process allows the system to adapt to different smartphone placements while ensuring high recognition accuracy.
  • Hierarchical CNN Classifier: A layered classifier is introduced to improve recognition accuracy and allow flexible scalability.
• Technical Implementation:
  • Main Steps:
    1. Calibration: Transforming IMU data into the human body coordinate system.
    2. Gesture Detection: Sliding window filtering of IMU data to capture gestures.
    3. Gesture Recognition: Classifying gestures using a hierarchical tree-based CNN model.
  • Preprocessing techniques (e.g., low-pass filtering) were applied, and model parameters were dynamically adjusted based on a brief four-step user calibration process.

Research Outcomes

• Specific Results:
  • User Study 1: Through target acquisition and confirmation tasks, users were found to easily distinguish 8 directional movements and 2 rotational directions. "Quick return" was identified as the optimal action confirmation technique.
    • User-Accepted Gestures: 8 movement gestures, with forward/backward movements replaced by forward-leaning/backward-leaning movements due to social considerations.
  • System Performance Evaluation: Gesture detection accuracy reached 97.5%, with a false positive rate of only 0.1% during daily use and a misclassification rate of <3%.
  • User Study 2: Tested in real-world and VR scenarios, HulaMove reduced interaction time by 41.8% (compared to traditional touchscreens) and significantly enhanced user immersion in VR environments.
• Comparative Advantages:
  • Unlike traditional hardware-dependent methods, HulaMove requires no additional hardware, making it highly versatile.
  • Compared to other interaction methods, its "vision-free, hands-free" feature is particularly suitable for specific work scenarios or enhancing immersive experiences.
• Limitations and Future Directions:
  • The design space for waist interaction remains underexplored (e.g., multi-layer circular regions, additional speed or motion trajectory dimensions).
  • Validation of feasibility needs to be extended to more diverse daily interference scenarios.
  • Mobile algorithm deployment requires further optimization (e.g., improving real-time performance and reducing power consumption).