Use Case Title
Zebra Crossing Monitoring – Detect Pedestrians Walking Across Marked Lanes
Team Members
Malhar Bipin Mhatre
Yash Anil More
Romit Tushar Meher
Tanishq Rakesh More
Mohammad Nisar Memon
Use Case Importance
Monitoring zebra crossings can improve pedestrian safety and reduce road accidents. Our solution detects pedestrians walking across marked lanes in real time, which can help traffic systems or drivers respond appropriately.
Data Collection and Annotation
We collected 600 images using mobile phone cameras at various zebra crossings under different lighting and traffic conditions.
Annotated Classes: “Pedestrian on Crossing”, “Pedestrian off Crossing”
Annotation Tool Used: Roboflow
Total Images: 600 (450 training, 150 validation)
Sample Images:
Model Training and Validation
We trained YOLOv8 Nano for efficient edge-device deployment.
Monitored Metrics: mAP@0.5, precision, and recall
We added 100 more images after observing lower recall for the “off Crossing” class and retrained the model to improve balance and accuracy.
Model Deployment and Demo Video
The model achieved 25 FPS on a mobile device and 87% accuracy in live testing.
Number of Models Deployed: 1 (YOLOv8 Nano) via the YOLOvX Android app
Demo Video: https://drive.google.com/file/d/1AifgwrPYYTBtf3l0JR-udSpUv7YtUVL-/view?usp=drive_link
Conclusion
Our solution showed reliable detection in live outdoor environments during the day.
We learned that detection accuracy drops at night or in low light, and plan to improve this with better lighting data or alternate camera inputs.
