Traffic surveillance video-based a unified road accident detection, classification, and vehicle tracking framework using an optimized spatio-temporal attention mechanism
- 1Department of Computer Science and Engineering, Sathyabama Institute of Science and Technology (Deemed to be University), Chennai, Tamil Nadu, India
- 2School of Computing, Sathyabama Institute of Science and Technology (Deemed to be University), Chennai, Tamil Nadu, India
Abstract
In recent days, vehicular accident detection has materialized because of the unique and intriguing application potential in the evolution of Intelligent Transportation Systems (ITS). To monitor and respond to the event of a fatal accident, real-world traffic surveillance footage requires continuous analysis. In this research, the road accident from the traffic surveillance video is detected, classified, and vehicle tracked using an efficient and innovative model. The surveillance videos are gathered from the CCTV Footage dataset, which is a publicly available dataset. The gathered videos are directly applied to the Spatio-Temporal Attention-based Adaptive Yolov9 with SE-InceptionNet and Deep SORT (AYSIDSORT) model for detecting, classifying and tracking traffic accidents very effectively. The developed network allows accurate detection and tracking of vehicles in real-time, which ensures the continuous traffic flow without losing moving objects. The variables of the proposed AYSIDSORT model are tuned by the developed Generation Probability-based Equilibrium Optimizer (GP-EO) to improve the detection and classification performance. The exact location of the accident is sent to the appropriate entities using an Internet of Things (IoT) platform to provide a timely rescue for the affected individual. Here, the proposed framework acts as a computer vision-based accident detection and tracking framework with potential IoT integration to find the exact location of the accident, which helps to provide a timely rescue for the affected individual. The effectiveness of the offered approach is validated with various baseline detection models. The accuracy of the suggested AYSIDSORT reaches to 95% in the 100th epoch, whereas there is a slight variation in the accuracy values when the batch size is varied from 4 to 128. The outcomes show that the designed framework is better than existing methods.
Related
- A deep-learning based gait classification and anomaly detection framework for healthcare surveillance
- A WHO CNS5-Aligned Classification Framework for Population-Based Surveillance of Brain and Other Central Nervous System Tumors in the United States
- Cross-corpus consensus feature selection for language-invariant Parkinson’s disease detection: a dual-stream acoustic and spectrogram-attention framework
- Designing an attention-based approach for RF fingerprinting in drone detection and classification
- An integrated wireless sEMG framework with Bayesian-optimized machine learning for multi-task locomotor classification and neuromuscular characterization
BibTeX
@article{devi2026traffic,
title = {{Traffic surveillance video-based a unified road accident detection, classification, and vehicle tracking framework using an optimized spatio-temporal attention mechanism}},
author = {Chokkakula Devi and Gowri S.},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0357111},
url = {https://doi.org/10.1371/journal.pone.0357111},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Traffic surveillance video-based a unified road accident detection, classification, and vehicle tracking framework using an optimized spatio-temporal attention mechanism
AU - Chokkakula Devi
AU - Gowri S.
PY - 2026
DA - 2026/10/15
JO - PLOS One
DO - 10.1371/journal.pone.0357111
UR - https://doi.org/10.1371/journal.pone.0357111
AB - In recent days, vehicular accident detection has materialized because of the unique and intriguing application potential in the evolution of Intelligent Transportation Systems (ITS). To monitor and respond to the event of a fatal accident, real-world traffic surveillance footage requires continuous analysis. In this research, the road accident from the traffic surveillance video is detected, classified, and vehicle tracked using an efficient and innovative model. The surveillance videos are gathered from the CCTV Footage dataset, which is a publicly available dataset. The gathered videos are directly applied to the Spatio-Temporal Attention-based Adaptive Yolov9 with SE-InceptionNet and Deep SORT (AYSIDSORT) model for detecting, classifying and tracking traffic accidents very effectively. The developed network allows accurate detection and tracking of vehicles in real-time, which ensures the continuous traffic flow without losing moving objects. The variables of the proposed AYSIDSORT model are tuned by the developed Generation Probability-based Equilibrium Optimizer (GP-EO) to improve the detection and classification performance. The exact location of the accident is sent to the appropriate entities using an Internet of Things (IoT) platform to provide a timely rescue for the affected individual. Here, the proposed framework acts as a computer vision-based accident detection and tracking framework with potential IoT integration to find the exact location of the accident, which helps to provide a timely rescue for the affected individual. The effectiveness of the offered approach is validated with various baseline detection models. The accuracy of the suggested AYSIDSORT reaches to 95% in the 100th epoch, whereas there is a slight variation in the accuracy values when the batch size is varied from 4 to 128. The outcomes show that the designed framework is better than existing methods.
SN - 1932-6203
ER -