Simulation-Based AI Prototype for Accident Detection and Emergency Intervention

View at publisher Summarise with VIRAT
Authors: Sharmila K., Bhavani Y., Sai Ram Tej K.
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Road traffic accidents represent a critical public health crisis, causing over 150,000 fatalities annually in India — manypreventable with timely intervention within the golden hour. Manual emergency reporting fails when victims are unconscious or bystanders are absent, and sequential contact with hospital and police services introduces compounding delays. This paper presents AccidentAlert, a simulation-based AI prototype that automates the complete emergency response pipeline on a standard Android smartphone without external hardware. The victim module provides a labelled SIMULATE ACCIDENT button that initiates a 30-second cancellable countdown, preventing false positives while ensuring minimal delay. A native Kotlin Random Forest classifier, running entirely on-device without network dependency, evaluates five features — vehicle speed, impact force, acceleration drop, vehicle type, and weather condition — to classify accident severity into Low, Medium, or High. The ensemble of 100 decision trees, trained on a 1,000 record synthetic dataset using an 80:20 train-test split, achieves 91.2% overall accuracy and outperforms a single Decision Tree baseline by 12.7 percentage points, with particularly strong High Severity precision of 93.4% — critical for appropriate emergency resource allocation. Upon classification, a structured accident record containing victim profile, GPS coordinates, severity label, and confidence score is written to Firebase Realtime Database and simultaneously dispatched via Firebase Cloud Messaging to dedicated Hospital and Police dashboards, replacing sequential call-chains with parallel dispatch. Responders can accept or reject cases and initiate one-tap Google Maps navigation to the victim. Average end-to-end notification latency across 50 field tests was 2.4 seconds, with a maximum of 4.1 seconds. All 15 defined system test cases passed successfully. AccidentAlert demonstrates a practical, scalable, and low-cost approach to reducing emergency response time in Indian road conditions, establishing a reproducible software baseline for future sensor-integrated and hybrid deployment architectures.Keywords: Accident Detection, Random Forest, Android, Firebase, Emergency Response, GPS Tracking, Severity Classification, Real-time Notification, Golden Hour

Fields of Study & Subjects

IoT and GPS-based Vehicle Safety Systems Traffic and Road Safety Energy and Environmental Systems

Publication details

Journal / source
International Scientific Journal of Engineering and Management
Volume
05
Issue
03
Pages
1–9
Publication date
31 March 2026
Language
EN