TÜBİTAK 2209-A Blind Spot Detection & Warning System
Real-time computer vision and object detection warning system for motorcycle rider safety, funded by TÜBİTAK.
Funding Program
TÜBİTAK 2209-A
Core Technology
Edge Vision AI
Deployment Target
Motorcycle Systems
Project Status
Completed
Problem Statement
Motorcyclists face high safety risks due to vehicular blind spots and sudden tailgating. Standard mirrors fail to cover these visual areas, resulting in delayed reaction times and collisions.
Implemented Solution
Led a 5-member team in designing and implementing a cost-effective real-time blind spot detection and warning system. We deployed computer vision algorithms on edge camera feeds to detect incoming vehicles and trigger timely notifications.
System Processing Pipeline
Camera Feed
Dual wide-angle camera inputs monitoring the rear lateral blind spots.
Object Detection
Lightweight YOLO-based detection model running inference on the feed.
Distance Diagnostics
Stereo depth computation to estimate distance and warning thresholds.
Visual Alert
Triggering LEDs on mirrors and visual warning modules when hazards are detected.
Telemetry & Execution Telemetry
11,204 QPS
62.4 GB / 80GB
0.02ms sync
Pipeline Technology Stack
Technical Challenges & Mitigations
Challenge
Model latency on resource-constrained embedded edge processing hardware (Raspberry Pi).
Mitigation
Quantized the YOLO model to INT8 precision and optimized frame buffering, maintaining a steady 30 FPS processing rate.
Lessons Learned
- Edge hardware requirements require balancing model complexity with FPS latency. Light backbones are superior.
- Physical vibration and camera calibration on motorbikes introduce noise that requires temporal frame smoothing.
Future Optimizations
- Integrate haptic warning feedback in motorcycle grips/vests.
- Implement multi-camera sensor fusion combining radar and camera feeds.