Automating Composition of Origin-Destination Flows of Intersections Based on UAV Data

Abstract

With the exponential development rate of UAV and computer version technologies, vast and sophisticated data on traffic is now available. Spatial and temporal data, including speed and other parameters of trajectory data, can be captured. Likewise, there are unsupervised clustering algorithms in the domain of machine learning that can group data points into clusters based on their inherent similarities without using labeled data. Algorithms such as GMM, DBSCAN, and HDBSCAN identify patterns and structures within the dataset, allowing for the discovery of natural groupings. Leveraging data from UAVs and these clustering algorithms, this paper aims to develop an effective and efficient methodology to automate the extraction of Origin-Destination (OD) flows of different types of intersections. A new custom-made intersection OD flow automation method called IODF is introduced, along with the deployment of DBSCAN, HDBSCAN, and GMM algorithms for clustering intersection trajectories, leading to the automatic extraction of OD flows. The results demonstrate that all four methods performed effectively in extracting OD flow for various intersection types.