Abstract :

Traffic Accidents Remain A Critical Global Challenge, Especially In Rapidly Expanding Smart Cities Where Dense Traffic Flows, Heterogeneous Road Structures, And Diverse Environmental Conditions Amplify Safety Risks. Traditional Accident Detection Mechanisms Depend Heavily On Manual Reporting, Delayed Responses, Or Sensor-based Systems With Limited Coverage. Recent Advances In Artificial Intelligence—particularly Deep Learning—offer Promising Solutions Capable Of Real-time Accident Identification From Video Feeds And Intelligent Transportation Infrastructures. This Study Proposes A Deep Learning Ensemble Framework Combining I3D Inflated 3D ConvNets, Optical Flow–based Motion Extraction, And ConvLSTM2D Spatiotemporal Models To Detect Accidents From Live Surveillance And Dashcam Streams. The Approach Takes Direct Reference From The High-performance Architecture Described In The Uploaded Study, Where Trainable Two-stream I3D + ConvLSTM2D Models Achieved 87% Mean Average Precision (MAP) And Robust Accuracy Across Varied Scenarios Including Rear-end, T-bone, And Frontal-impact Collisions. In The Proposed System, RGB Frames Capture Spatial Cues While Optical Flow Highlights Motion Irregularities, Enabling Strong Temporal Modelling Even In Complex Or Cluttered Environments. Ensemble Fusion Enhances Classification Reliability, While The Model’s Lightweight Design Supports Deployment On Edge IoT Devices For Real-time Urban Monitoring. Experimental Insights From The Reference Research Demonstrate That Integrating Multi-stream Features Significantly Improves Accident Detection Accuracy, Outperforming Traditional Models Such As SSD, DETR, DenseNet-Transformer, And Baseline CNN/LSTM Architectures. This Work Contributes An Efficient, Scalable Accident Detection Solution Ideal For Smart City Environments, Capable Of Accelerating Emergency Responses, Reducing Congestion Impacts, And Enhancing Public Safety Through Predictive Urban Mobility Intelligence. Keywords: Traffic Surveillance, Accident Detection, Action Recognition, Smart City, Autonomous Transportation, Deep Learning.

Published:

23-2-2026

Issue:

Vol. 26 No. 2 (2026)

Page Nos:

147-155

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite