Application of deep learning to improve the performance of collision recognition algorithms in 3D space

Abstract

The prevention and detection of automotive accidents have been the focus of numerous studies, with many aiming to identify potential accident-causing objects or analyzing accident statistics. This research introduces a system designed to detect random accidents by collecting data from nearby vehicles and processing it using machine learning tools to identify potential accidents. The objective is to analyze road behavior and identify vehicles deviating from normal norms. The study utilizes the SUMO (Simulation of Urban Mobility) traffic simulator to simulate vehicle movement and gather information on their positions and sleep patterns. A total of 100 vehicles were simulated on a 3000-meter road segment using SUMO, with various vehicle types to mimic real-world traffic. The system utilizes the DBSCAN unsupervised clustering algorithm to detect anomalies, showing its effectiveness in identifying accidents. Results demonstrate that anomalies increase after an accident, indicating blocked roads or significant traffic increase in alternative lanes. Automated accident detection is crucial for traffic management systems, aiding in the avoidance of future incidents and enabling authorities to reopen roads promptly. This study highlights the potential of analyzing traffic behavior through vehicle positions and speeds, with abnormal activities serving as potential threats to nearby drivers. The findings underscore the importance of incorporating machine learning techniques into traffic management systems to enhance road safety and efficiency. In conclusion, this research demonstrates the feasibility and efficacy of utilizing machine learning algorithms for automated accident detection in traffic management systems. By leveraging advanced technologies, such as simulation and clustering analysis, road safety can be significantly improved, ultimately leading to safer and more efficient transportation networks.