LaserShoes: Low-Cost Ground Surface Detection Using Laser Speckle Imaging

Document Title

LaserShoes: Low-Cost Ground Surface Detection Using Laser Speckle Imaging

Document Information

• Subject Area: Ground surface detection in wearable devices and human-computer interaction (HCI)
• Keywords: Smart shoes, laser speckle imaging, surface recognition, texture recognition, context-aware computing, real-time classification, motion analysis

Research Background and Problem

• Problem or Challenge:

  • Current surface detection technologies often rely on visual imaging or foot dynamics, which may be influenced by user motion characteristics or health conditions.
  • Traditional cameras cannot capture subtle surface textures and often require lenses, which may compromise privacy.
  • Developing a reliable ground detection system for dynamic environments and diverse surface types remains a technical challenge.

• Significance: Ground surfaces carry rich contextual information and are applicable to activity recognition, health monitoring, and context-aware computing in various scenarios, such as enhanced motion analysis and hazardous surface detection.

• Motivation and Related Work:

  • Existing work, such as smart shoes and ground texture detection technologies, faces challenges such as high equipment costs and low robustness in dynamic environments.
  • Laser speckle imaging technology has been applied in other fields (e.g., medical blood flow detection) but has not been fully explored for ground detection.

Solution

• Method or Solution:

  • Proposed a low-cost ground surface detection system named LaserShoes, which captures ground surface textures using laser speckle imaging technology.
  • The system consists of a shoe-mounted sensing component and a leg-mounted processing component for portable and real-time ground type detection.
  • Developed an end-to-end detection process, including data preprocessing and deep learning classification.

• Innovations:

  • Applied laser speckle imaging to dynamic foot-ground contact detection scenarios, avoiding interference from gait or visual effects in traditional methods.
  • Excluded blurred images during dynamic motion in the preprocessing stage and performed real-time inference using a deep learning classification model.

• Implementation Steps and Key Technologies:

  1. Hardware Design:
     • Utilized a 520nm laser emitter and CCD image sensor to construct the detection component.
     • Raspberry Pi Zero 2 W was used for data collection and processing.
     • Designed optical and mechanical structures to maintain the stability of the detection component.
  2. Data Preprocessing:
     • Converted images to grayscale, identified foot-ground contact images, cropped images, and excluded blurred or unclear texture images.
  3. Deep Learning Classification:
     • Used the ResNet-18 model for texture classification, with input being the preprocessed clear images.
  4. Real-Time Inference:
     • Optimized processing and classification steps to achieve real-time feedback.

Research Outcomes

• Specific Results:

  • The LaserShoes system demonstrated high classification accuracy for ground surface types in user studies.
  • Achieved an average classification accuracy of 86.93% (same-user data) and 80.57% (cross-user model) across 24 ground surface types.

• Advantages Over Existing Solutions:

  • Successfully classified surfaces under different gaits and environments (e.g., wet or icy surfaces), demonstrating system robustness.
  • Distinguished visually similar surfaces, such as light-colored wooden floors and plastic floors.

• Experiments and Evaluation Results:

  • Experiments covered various indoor and outdoor ground surfaces, validating the system's performance under different lighting conditions and on dynamic surfaces (e.g., sandy particles).
  • Achieved real-time classification, though the Raspberry Pi platform showed slight performance limitations.

• Limitations and Future Directions:

  1. Limitations:
     • High power consumption limits prolonged system use.
     • Current implementation cannot predict hazardous surfaces in advance.
     • Weak detection performance on transparent surfaces (e.g., glass) and loose materials (e.g., grass).
  2. Future Directions:
     • Optimize hardware design to reduce device size and improve stability.
     • Enhance detection performance for transparent and loose materials.
     • Introduce optical filters or high dynamic range sensors to address strong light interference.
     • Implement airborne state detection to enable early warning functionality.

Conclusion

The LaserShoes study demonstrates the innovative application of laser speckle imaging technology in ground detection and validates its feasibility and robustness in various scenarios. Future improvements include optimizing energy consumption, enhancing adaptability to lighting conditions, and improving the accuracy of dynamic surface detection. This work opens new possibilities for the application of wearable smart devices in context-aware computing and activity analysis.