Fusing Visual Quantified Features for Heterogeneous Traffic Flow Prediction

Abstract

This paper presents a novel traffic flow prediction method emphasising heterogeneous vehicle characteristics and visual density features. Traditional models often overlook the variety of vehicles, resulting in inaccuracies. The proposed method utilises visual techniques to quantify traffic features, such as mixed flow and vehicle accumulation, enhancing dynamic density estimation and flow fluidity. We introduce a spatio-temporal prediction model that integrates various data types, capturing complex dependencies and improving accuracy. This research advances traffic flow prediction by considering the diverse nature of vehicles and leveraging visual data, offering valuable insights for intelligent transportation systems. Experimental results demonstrate the superiority of this approach over conventional methods, especially in capturing traffic flow fluctuations.