Introduction
Smart wearables for the Blind and Visually Impaired (BVI), such as the VIS⁴ION system, are designed to provide real-time situational awareness. However, these devices are currently hampered by a "triple threat" of constraints: limited local computational power, narrow fields-of-view from single-camera setups, and the high latency/low bandwidth of 4G/LTE networks. This paper investigates how 5G millimeter-wave (mmWave) and mobile edge computing (MEC) can overcome these barriers by offloading heavy processing tasks to the network edge.
Methodology
The researchers developed a comprehensive evaluation framework combining computer vision analysis and high-fidelity wireless network simulations.
Key methodological pillars include:
- Vision Evaluation: Assessing how resolution (from WVGA to 2.2K) and compression impact the accuracy of the YOLOv5s object detection model.
- Wireless Simulation: Utilizing the ns-3 simulator with ray tracing on 3D models of Manhattan to realistically model 5G mmWave signal propagation and blockage.
- Adaptive Offloading: Implementing a scheme that dynamically adjusts video resolution and the number of active camera feeds based on real-time network throughput and delay.
- NYU-NYC StreetScene Dataset: Introducing a new benchmark dataset captured from a pedestrian's perspective in New York City, featuring high-resolution video and distance-based annotations.
Key Insights and Results
The study defines a 100ms "safety budget" for total system latency, based on the duration of a human saccade. The findings highlight the transformative potential of 5G:
- Resolution vs. Distance: High-resolution 1080p feeds, enabled by 5G's bandwidth, double the reliable detection range for pedestrians from 6 meters to 12 meters compared to standard WVGA feeds.
- Throughput and Latency: 5G mmWave successfully supported four simultaneous 1080p camera feeds at 26 Mbps each. The median round-trip time (RTT) was reduced to 15ms (vs. 37ms for LTE), allowing the system to stay within the safety budget 75% of the time.
- Edge Processing Advantage: Offloading to an edge server with high-end GPUs allows for processing 1080p video at ~50 fps, whereas local mobile processors struggle to exceed 6 fps at the same resolution.
- Hybrid Reliability: While mmWave is sensitive to blockages, the proposed adaptive scheme ensures 100% availability by falling back to LTE or local processing during 5G outages.
Conclusion
By leveraging the massive bandwidth and low latency of 5G mmWave, smart wearables can achieve 360-degree situational awareness with significantly higher detection ranges than current solutions. Future work will focus on improving mmWave robustness against human body blockage and optimizing models for compressed video streams to further enhance edge-computing performance.
Summary updated