PathFinder: Designing a Map-less Navigation System for Blind People in Unfamiliar Buildings

Title of the Paper

PathFinder: Designing a Map-less Navigation System for Blind People in Unfamiliar Buildings

Paper Information

• Research Domain: Map-less navigation technology, design and evaluation of assistive devices for blind individuals
• Keywords: Visual impairment, orientation and mobility, intersection detection, signage recognition, map-less navigation

Research Background and Problem

• Identified Challenges:
  • Blind individuals face difficulties navigating unfamiliar buildings independently, requiring extensive familiarity with the environment or relying on assistance from others.
  • Existing navigation systems rely on pre-built maps, which are time-consuming and costly to construct, limiting the effective coverage of these systems.
• Significance:
  • Assisting blind individuals in safely navigating unfamiliar environments independently can significantly improve their quality of life and confidence while reducing the societal costs of accessible environment technologies.
• Research Motivation and Related Work:
  • Many navigation systems utilize static route maps and positioning devices (e.g., BLE beacons or LiDAR maps) to assist blind individuals, but these technologies face scalability limitations.
  • The study aims to explore map-less navigation technology to reduce map construction costs and investigate the actual needs and design recommendations for blind users through participatory design research.

Solution

• Proposed Solution:
  • Designed and implemented a map-less navigation system called PathFinder, utilizing a suitcase-style robot combined with intersection detection and signage recognition technologies to assist blind individuals in navigating unfamiliar buildings.
• Innovations:
  • Does not rely on pre-built maps, instead using real-time environmental information (e.g., intersections and signage) to guide users.
  • Introduced a "Take-me-back" feature to return to the navigation starting point, enhancing the overall user experience.
• Implementation Steps:
  1. Conduct participatory design research to analyze blind users' needs for navigation systems (e.g., reliance on intersection and signage information).
  2. Develop a preliminary prototype, including intersection detection and signage recognition modules:
     • Intersection detection uses 360° LiDAR and SLAM to construct real-time maps, extracting passage directions and intersection shapes.
     • Signage recognition employs object detection (YOLOv5) and OCR technologies to distinguish directional and textual signage, providing relevant feedback to users.
  3. Improve user interface and feedback interaction:
     • Update intersection direction feedback voice (e.g., using "left" and "right" instead of clock directions).
     • Optimize system button layout for quick activation of signage recognition.
     • Add a return function to enhance user convenience.
  4. Conduct adaptive experiments to fine-tune module performance and carry out quantitative and qualitative user testing.

Research Outcomes

• Specific Results:
  • PathFinder effectively assists blind individuals in navigating to their destination, boosting user confidence and reducing cognitive load.
  • The system expands the navigable environments for blind individuals in scenarios where top-line systems (using pre-built maps) are unavailable.
• Advantages Comparison:
  • Compared to conventional navigation aids (e.g., guide dogs or canes):
    • PathFinder provides real-time environmental feedback, significantly enhancing user perception.
  • Compared to systems based on pre-built maps:
    • PathFinder covers a broader range of use cases, reducing reliance on map construction.
• Experimental and Evaluation Results:
  • Conducted seven user tests across two routes with different attributes:
    • Intersection detection achieved an overall accuracy rate of 56%, though some results require improvement (e.g., false detections caused by glass bridges).
    • The signage recognition module successfully identified 27 effective signs out of 62 activations.
    • User feedback indicated high applicability of PathFinder, with an average System Usability Scale (SUS) score of 85.25.
  • Compared to top-line systems, task completion time was slightly longer, but users reported greater control and flexibility.
• Limitations and Future Directions:
  • Limitations:
    • The robot's intersection detection performance is lower in crowded or high-traffic environments.
    • Open spaces (e.g., lobbies or wide corridors) pose challenges for navigation algorithms.
    • The design of the segmented device limits portability and battery life.
    • Users require short-term training to learn system interactions, with a lack of long-term usage testing.
  • Future Directions:
    • Optimize intersection detection algorithms to improve accuracy in more complex environments.
    • Enhance signage recognition classification capabilities, such as directly distinguishing between directional and textual signage.
    • Develop more portable form factors and explore multifunctional device support (e.g., wearable technology).
    • Enable users to autonomously construct partial maps or annotate scenes to expand navigable areas.