Accident Detection and Criticality Assessment System

Precision
Recall
Confusion Matrix
F1-Score
Image for explaining second feature

Inspiration

Accidents pose challenges like delayed responses, inefficient recovery, and lack of critical assessments. This project addresses these gaps with an integrated and efficient accident management system.

What it does

The system detects accidents using a fine-tuned YOLOv11 model, saves accident frames, and analyzes them with a vision-language model to assess criticality. It ensures rapid GPS-based response coordination, estimates damage costs, and connects users to repair services, providing end-to-end accident management.

How we built it

Fine-tuning YOLOv11 for diverse accident scenarios.
Accurate criticality assessment using vision models with accident frame descriptions securely stored in Firebase for a safe environment.
Seamless integration of various features like GPS, damage estimation, and service connections.

Challenges we ran into

One major challenge was the YOLOv11 model misclassifying red cars as accidents. To address this, we expanded our dataset by annotating additional images of red cars in various non-accident scenarios. This improved the model's ability to differentiate between normal and accident cases, enhancing detection accuracy.

Accomplishments that we're proud of

Successfully fine-tuned YOLOv11 for real-time accident detection.
Developing an end-to-end system that not only detects accidents but also facilitates recovery.
Enhancing safety and efficiency in post-accident processes.
Implementing secure storage of accident frame descriptions in Firebase, ensuring a safe environment for sensitive data.

What we learned

The importance of balancing accuracy and efficiency in accident detection.
Integrating multiple technologies like vision-language models, yolo-v11 and ai agent.
The challenges of handling live data streams efficiently for accident detection and response.

What's next for Accident Detection and Criticality Assessment System

Expanding datasets to improve YOLOv11 performance for edge cases.
Adding real-time notification features for emergency responders.
Exploring partnerships with insurance companies for automated claims processing.
Improving scalability for deployment in smart city infrastructures.