DEEP LEARNING–BASED ACCIDENT DETECTION IN INTELLIGENT TRANSPORT NETWORKS
Anil Laxmanrao Wakekar, Kottala Sri Yogi, Vishal Anil Aher, Harshada Bhushan Magar, Ashutosh Kulkarni, Deepika Sharma
Abstract: The rapid development of intelligent transport systems has contributed greatly to traffic density thus timely and accurate detection of accidents is one of the key demands in improving road safety and emergency response. Older sensor and rule based methods are characterized by slow detection, inadequate scalability and resistance to more complex traffic conditions. In order to tackle such issues, this paper proposes an accident detection system that is based on deep learning and is applicable in the context of intelligent transport systems. The main aim of the study is to design and perform the evaluation of strong deep learning frameworks with the ability to automatically identifying traffic accidents on heterogeneous types of traffic data with high accuracy and low latency. The framework is based on four deep learning approaches, which include Convolutional Neural Networks (CNNs) to extract spatial features of traffic images, Long Short-Term Memory (LSTM) networks to model temporal traffic dynamics, a CNNLSTM hybrid architecture to identify accidents on a spatiotemporal scale, and a Transformer-based attention system to learn long-range dependencies and be mindful of context. These models are supervised trained and evaluated on traffic video and sensor datasets preprocessed and inputted into them. The experimental findings indicate that the CNN model has a higher accident detection rate of 91.3% compared to LSTM model with 93.8% because it is able to extract and capture the temporal patterns. The CNN LSTM hybrid has a high accuracy of 96.4% and a low false alarm rate of 3.1 as compared to individual models. The Transformer-based model has the highest performance with an accuracy of 97.9 %, 98.2 % and lower detecting latency than the current methods. These results prove that accident detection using deep learning has the potential to greatly improve reliability, responsiveness, and scalability in intelligent transport networks.
Keywords: Accident detection; Intelligent transport networks; Traffic safety; Spatiotemporal analysis; Real-time monitoring
DOI: 10.24874/PES08.03.006
Recieved: 12.05.2026 Revised: 26.06.2026 Accepted: 14.07.2026
UDC:
Reads: 11 
