A Study on Temporal and Spatial Distribution Characteristics of Traffic Flow Based on YOLO Model

Abstract

With the increasing urban population urban traffic behaviors are becoming increasingly complex Therefore for urban planners how to analyze the characteristics of traffic behaviors is an extremely important issue In order to further study on temporal and spatial distribution characteristics of traffic flow in the microscopic scale this study uses the most densely distributed roadside monitor video as the data source Then through the YOLOv3 deep learning framework to automatically detect three kinds of vehicles and combined with the multiple object tracking algorithm DeepSORT to extract long-term vehicle flow changes and vehicle trajectories of each road section Next correlate with the aerial image calculate the homography matrix to carry out the projection transformation of the two coordinate system in order to get the microscopic traffic characteristics in real-world coordinates such as instantaneous speed instantaneous acceleration and lateral offset distance etc The other part uses space syntax to quantify road network topology indexes such as global relative integration (Rn) local relative integration (R3) and road network Choice Finally through the correlation analysis this study obtained similar results as the previous studies The road width and the global relative integration can effectively predict the distribution of vehicle flow and the time-varying characteristics of vehicle flow will be affected by the level of the road network; In addition the spatial distribution and correlation of microscopic traffic characteristics are also used to infer the main factors that affect the differences in urban traffic characteristics

Date of Award	2020
Original language	English
Supervisor	Han-Liang Lin (Supervisor)