Enhancing safety of micro-mobility and powered mobility devices using YOLOv12-based real-time obstacle detection
The increasing use of micro-mobility devices (MMDs) and powered mobility devices (PMDs) has improved personal mobility but raised safety concerns, particularly for elderly individuals and persons with disabilities who may have limited reaction times. This paper presents a real-time obstacle detectio...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Proceeding Paper |
| Language: | en en |
| Published: |
IEEE
2025
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/127111/7/127111_Enhancing%20safety%20of%20micro-mobility.pdf http://irep.iium.edu.my/127111/8/127111_Enhancing%20safety%20of%20micro-mobility_Scopus.pdf http://irep.iium.edu.my/127111/ https://ieeexplore.ieee.org/document/11233477 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The increasing use of micro-mobility devices (MMDs) and powered mobility devices (PMDs) has improved personal mobility but raised safety concerns, particularly for elderly individuals and persons with disabilities who may have limited reaction times. This paper presents a real-time obstacle detection and alert system to enhance user safety without compromising manual control. The system integrates a Raspberry Pi 4 with a camera, MPU-6050 accelerometer, and buzzer, utilizing the YOLOv12 object detection model and DeepSORT tracking algorithm. Trained on 5,000 COCO images with 36,335 instances, the model achieved a precision of 0.601, a recall of 0.386, a mAP@50 of 0.419, and mAP@50–95 of 0.289. The model was converted to the NCNN format for lightweight deployment, enabling average inference at 214.3 ms per frame. Field tests across environments with minimal, moderate, and high obstacle densities validated the system’s real-time tracking, with DeepSORT reducing duplicate detections by over 40%. Audible alerts reliably notified users upon hazard detection. This work demonstrates the viability of embedding AI-powered safety systems on low-cost hardware for mobility applications. The prototype effectively enhances situational awareness, offering a scalable solution to reduce preventable accidents and support vulnerable user populations. Future work will address lighting variability and braking integration. |
|---|