The Authenticity Evaluation Model Of Traffic Flow Simulation Based On Dictionary Learning

In recent years,traffic animation is receiving considerable attention and has been widely used in various applications.Researchers have proposed several models based on the macroscopic and microscopic characteristics of traffic flows.In some applications like urban traffic management,authenticity of the crowd animation is highly required.However,how to evaluate these models is rarely considered.This thesis presents a novel evaluation model for traffic simulation results using dictionary learning.Our model borrows some ideas from realistic traffic flow animation using texture synthesis,and takes traffic flows as signals.Given a pool of real traffic data,we first extract some low-dimensional characteristic signals from them.Then,we extract their structural characteristic signals through a dictionary learning algorithm.After that,we evaluate each traffic simulation method by comparing the results from a simulation method with real data in the dictionary.Our thesis provides a novel quantitative evaluation method and a feedback correction mechanism for traffic simulation,which not only promotes the development of this research direction but also plays an active role in the governance of urban transport.In addition,the dictionary learning-based algorithm can also be employed in statistical similarity measure for aggregate crowd dynamics.