VisuPath

Arduino® UNO™ Q 4GB

Modulino™ Vibro

Modulino™ Distance

Arduino App Lab

<html><body>Navigating crowded or unfamiliar environments can be challenging for people with visual impairments. Traditional mobility canes are effective at detecting obstacles and changes in ground level within the cane’s physical reach, but they generally cannot identify an object before contact, describe what it is, or warn the user about hazards farther ahead. VisuPath explores how edge AI, computer vision, proximity sensing, vibration, and spoken alerts can provide an additional layer of environmental awareness. The device is designed as a compact attachment that can be mounted onto a preexisting mobility cane, so that users can continue using the cane they already know and trust. VisuPath is portable, battery-powered, and modular. Its primary object-detection and vibration functions can operate without cellular service. <h3>Implementation</h3>On-Device Object Detection VisuPath is built around the Arduino UNO Q. A USB camera captures the environment ahead while an on-device YOLOv8 model, converted to TensorFlow Lite through the Qualcomm AI Hub pipeline, detects navigation-relevant objects such as people, bicycles, vehicles, chairs, traffic lights, and stop signs. Detections are filtered by confidence and non-maximum suppression to remove weak or overlapping results. The system prioritizes relevant hazards and identifies whether an object appears to the left, ahead, or to the right. It can generate alerts such as: <ol><li>“Person detected ahead, nearby.” </li><li>“Bicycle detected on the left.” </li><li>“Car detected on the right, farther away.” </li></ol>Distance Detection and Estimation VisuPath uses two methods to determine obstacle proximity. The Arduino Modulino Distance sensor directly measures nearby objects in front of the cane, while the camera pipeline estimates proximity using the size of each object’s bounding box, classifying it as “very close,” “nearby,” or “farther away.” The system also compares detections across consecutive frames and uses approximate real-world object dimensions and the camera’s field of view to estimate distance, track movement, and determine whether an object is approaching or moving away. Haptic Feedback The Arduino Modulino Vibro provides immediate physical feedback based on measurements from the distance sensor. No vibration is produced when an obstacle is beyond approximately 1100 millimeters. The vibration becomes stronger as the obstacle gets closer and reaches maximum intensity at approximately 300 millimeters or less. This provides an intuitive indication of proximity without requiring the user to listen to continuous spoken alerts. Smartphone Alerts and Text-to-Speech VisuPath sends detection information from the Arduino UNO Q to a connected phone or browser through WebSockets. A Node.js server receives alerts from the board and forwards them to a text-to-speech client. The phone can announce the detected object, its direction, and its approximate proximity. Object detection and message transmission run as separate processes. A small queue keeps recent alerts and discards older ones when necessary, helping prevent delayed messages from building up. The board-side client also attempts to reconnect automatically if the connection is interrupted. Phone-Based Scene Understanding The project also uses a vision-language model on the phone to provide broader scene explanations. Instead of only announcing individual objects, the VLM can describe the overall situation, such as a crowded path, an entrance surrounded by people, or an obstacle partially blocking the walkway. This phone-based scene-understanding feature complements the core system. The Arduino board handles low-latency object detection, proximity sensing, and vibration feedback, and the phone model provides more detailed scene descriptions. <h3>Results</h3>The current VisuPath prototype integrates on-device object detection, direct distance sensing, adaptive haptic feedback, and smartphone voice alerts to identify nearby obstacles and communicate their type, direction, and proximity in real time. The working prototype can: <ol><li>Capture a live camera feed </li><li>Run YOLOv8 object detection on the Arduino UNO Q </li><li>Filter for navigation-relevant objects </li><li>Identify whether an object is left, ahead, or right </li><li>Estimate general proximity from the camera image </li><li>Measure nearby distance with the Modulino Distance sensor </li><li>Increase vibration intensity as an obstacle gets closer </li><li>Prioritize important detections </li><li>Send alerts to a phone or browser </li><li>Convert received alerts into speech </li><li>Perform primary detection without cellular service </li></ol><h3>Potential future Improvements</h3><ol><li>Specialized Navigation Recognition: Additional models could be trained to recognize features that general-purpose object detection might not handle well </li><li>Validated Scene Understanding: The phone-based VLM could be developed into an optional feature that answers questions such as “Is the path clear?” or “Where is the nearest entrance?” </li><li>Personalization: Future versions could allow users to customize vibration strength, alert frequency, speech rate, distance thresholds, and which objects receive priority. </li><li>User-Centered Testing: The most important next step is testing with people who have visual impairments, orientation and mobility specialists, and accessibility researchers. Their feedback would guide decisions about device weight, mounting position, vibration patterns, alert frequency, comfort, privacy, and cognitive load. </li></ol></body></html>

VisuPath

Smart Mobility Cane: Enhancing orientation & spatial awareness for the visually impaired beyond the range of an ordinary cane.