Title Accident Identification and Collision Probability Estimation for Roadway Traffic: Applications of SVM and Random Forests
Year 2019
Degree Master
School National Cheng Kung University Department of Transportation and Communication Management Science
Author Huang, Po-Tsung
Summary        Currently, with the development and progress of communication technology, governments all over the world also vigorously promote regulations and security measures. They utilize the intelligent transportation systems (ITS) to not only improve safety, efficiency, and traffic problems, but also prevent traffic accidents. Through analyzing big data, distance, and other substantial data, ITS not only infers the risks of road safety, but also gives drivers the necessary promptings and cautions. Furthermore, the expectation of people for autonomous vehicles are also increasing because they hope that autonomous vehicles can be improved the risk prediction and accident prevention mechanisms to achieve driving safety.
       This study simulates the operating method of autonomous vehicles judging the external environment or moving objects from the angle of artificial intelligence. If there has moving objects classified as the potential threat of occurring traffic accidents, how we should define the collision probability between two cars. This study chooses support vector machine (SVM) and Random Forests as our methodology. Through capturing the specific features from the videos of traffic accidents, we acquire the change of every feature in different time periods before happening traffic accidents.
       After that, the features in different time periods are imported into two algorithms to train a high-accuracy model to identify the collision probability between two cars in different situations. Comparing the training results of two algorithms, we can know which algorithm has better performance, and adopt the algorithm with the best results as the high-accuracy model. In the future, this model can predict the collision probability in certain time period of video when we import a video’s features.
Hashtags
View count: 122