Research On Heterogeneous Traffic Flow Prediction Method Based On Visual Quantification

In the construction and development of smart cities,the process of intelligentization of road traffic plays a very important role.By analyzing and predicting traffic data,real-time traffic conditions in the road network can be understood in a timely manner,improving the ability to regulate traffic and handle sudden traffic incidents.With the vigorous development of big data and artificial intelligence technology,traditional methods cannot meet the demand due to the increasingly complex traffic data.Therefore,more and more research tends to use deep learning-related methods for traffic data analysis and prediction.At the same time,based on the characteristics of wide sampling range and rich data information,visionbased traffic quantification methods are increasingly favored.It can not only obtain microscopic information about road vehicles through image data but also obtain macro traffic indicators such as traffic flow,lane occupancy,and queue length.Therefore,this thesis quantifies traffic flow and traffic density indicators based on machine vision methods and uses spatial flow feedback algorithms to achieve heterogeneous traffic flow prediction of urban traffic networks.The main research content of this paper is as follows:(1)Unlike traditional methods that rely solely on data collected through ground induction coils,this paper uses visual detection technology to quantify traffic indicators.By utilizing computer vision algorithms,the proposed method can more comprehensively and accurately evaluate traffic flow characteristics.The method described in this paper enables the quantification of traffic features beyond the scope of traditional sensors,including traffic density features reflecting the level of road congestion and heterogeneous traffic features reflecting road composition.With quantified visual data,more detailed and accurate feature data can be obtained for subsequent traffic forecasting work in urban roads.(2)Most current spatiotemporal traffic flow prediction methods consider the impact of upstream traffic on downstream traffic,while neglecting the feedback effect of downstream traffic in congested conditions.This paper proposes a feedback traffic flow prediction method based on visual traffic features.In traffic flow theory,downstream traffic vehicles can affect traffic flow in multiple ways as they travel along the road.For example,when the road narrows downstream,such as in construction zones or bottlenecks,the space for vehicles to travel decreases,leading to an increase in traffic density and a decrease in traffic speed.This may cause a shockwave effect,exacerbating and spreading traffic congestion upstream,causing delays and reducing overall traffic flow.This paper proposes a feedback traffic flow prediction method based on visual traffic features to quantify macroscopic traffic flow indicators and flow feedback in density features in time series and spatial features.Experimental results using the STREETS dataset show that the proposed model outperforms state-of-the-art methods,particularly in predicting sudden changes in traffic flow,producing more accurate predictions during non-periodic peak traffic periods.(3)The paper addresses the issue of most traffic methods assuming a homogeneous ideal state for the entire traffic system when predicting traffic flow.A heterogeneous traffic flow prediction method based on visual density features is proposed in this paper.Heterogeneous traffic flow takes into account the varying impacts of different types of vehicles,pedestrians,and other factors on traffic flow prediction,rather than assuming homogeneous factors.For example,the effects of traffic flow changes are different for heterogeneous vehicle types due to differences in size and braking performance.The paper explores traffic flow prediction by using a heterogeneous traffic flow prediction model to obtain relevant features.Finally,comparative and ablation experiments demonstrate that the proposed method provides more accurate traffic flow predictions in mixed states involving large vehicles,non-motorized vehicles,pedestrians,etc.