DEEP LEARNING BASED TRAFFIC STATE ESTIMATION SYSTEM AND METHOD THEREOF

摘要

The traffic state estimation system includes multiple cameras and servers.Multiple cameras capture multiple index finger areas and vehicles on the aforementioned index finger areas to generate raw videos.The non index finger region is the area between adjacent index finger regions in multiple index finger regions.A deep learning based traffic state estimation model is stored on the server to estimate the traffic state in non detection areas.The server can communicate with multiple cameras.Multiple index finger regions each include at least a portion of the intersection.The traffic status in the non detection area is defined by the number of vehicles in the non detection area and the traffic density.The server utilizes video analysis algorithms to extract vehicles as objects from multiple original videos.The server tracks objects and generates vehicle trajectory information.The server extracts the features of the object and generates vehicle feature information.