Enhancing Traffic Safety with Advanced Machine Learning Techniques and Intelligent Identification

Read the full article See related articles

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Urban traffic safety is a critical concern due to the frequent lane changes and merges that create complex traffic flow patterns. Traditional methods, such as overhead video recordings, are commonly used to study these zones, but they come with high costs and limitations due to environmental factors like lighting and weather conditions. This work addresses the critical challenges in diversion and merging zones for urban traffic safety, with the primary objective of comprehensively tracking and identifying vehicles from video or image data. The process begins with generating traffic data using the SUMO (Simulation of Urban MObility) platform, which creates a simulated environment for urban traffic within synthetic road networks, taking into account traffic rules, signals, and other relevant factors. The generated data includes vital information such as vehicle IDs, coordinates, speeds, and road segments, providing a detailed representation of traffic dynamics. The next step involves utilizing the yolov8-deepsort framework to analyze driving behavior by accurately tracking and identifying vehicles in the simulated environment. This is followed by real-time risk assessments and the enhancement of traffic safety management. For conflict identification, CatBoost is employed due to its robustness and efficiency. To optimize model performance, CatBoost is further refined using Bayesian Optimization (CatBoost-BO), which fine-tunes the model's hyperparameters. Additionally, SMOTE is applied to address sample imbalance, resulting in a more balanced and accurate model. The model's performance is rigorously evaluated using metrics such as the confusion matrix, accuracy, recall, F1 score, and AUC-ROC, ensuring a comprehensive assessment. Furthermore, SHAP values are used to interpret the model, offering valuable insights into the factors contributing to safety risks. This interpretability is crucial for understanding and mitigating traffic conflicts. The study presents a practical and effective approach to improving urban traffic safety through advanced data analysis and machine learning techniques.

Article activity feed