Work

Real Time Accident Monitoring Using Vision Language Models

Year: 2025

Type: preprint

Abstract: Accidents are a major worldwide cause of mortality, with 5 lakh cases per year being reported in India, according to the Ministry of Road Transport and Highways. In order to overcome the constraints of low lighting, obstructions, and poor image quality, this research proposes a vision-language model for the real-time detection and classification of accidents from CCTV images. Using a dataset of 4,662 photos (both accident and non-accident categories), the system makes use of 20 cutting-edge vision-language models, such as metaclip-114-400m and CLIP-ViT-L-14-laion2B-s32B-b82K. The system doesn't require a lot of labeled training data to identify accident indicators like crashes or odd vehicle movement. Training on a diverse collection of photos strengthens the model, achieving 99% accuracy even under challenging circumstances. When traditional methods fail, this technology helps traffic management systems, expedites emergency response, and improves public safety.

Source: SSRN Electronic Journal

DOI: 10.2139/ssrn.5221673

Authors: Naga Raju challa, Bhuma Chandra Mohan, Pinniboina Bhavani, Olutayo Oyerinde, Kalapraveen Bagadi, Visalakshi Annepu, Lingisetty Naga Pavan Kumar

Institutions: University of the Witwatersrand, VIT-AP University, Vellore Institute of Technology University, Webster University

Language: English

Topic: Advanced Neural Network Applications

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

SDG: Sustainable cities and communities

Open Access status: green

OpenAlex ID: W4413514808

Cites:

Cited by:

Related to: 10