An end-to-end unified framework for assessing Turning Movements based Deep Neural Network

Abstract

Road traffic is a major issue in cities with increasing population. An Efficient traffic management system is required to address this issue. Turning Movement (TM) count is used for such purposes. TM acts as an input parameter for traffic analysis models and forecasting tools. This paper discusses the optimization of Turning Moments' count which can be utilized to predict road traffic accurately using an Artificial Neural Network (ANN) model. This is achieved by using path flow estimator to estimate traffic at hyperlink flows and Kalman filter to generate a dynamic model. Together the ANN model combined with a potent model with appearance constraint allows the model to consider special conditions and factors that affect the TM. Image capturing acts as the key concept to keep the count of TM more accurate and datasets are generated to save the data accordingly to keep features and size-match of the convoluted snapped of the vehicles. The representational model is used to solve matching ambiguities which can arise throughout dense traffic situations and occlusions and guarantees presenting consistency alongside a trajectory. The dynamic appearance model taken into consideration of the external factors that may affect the traffic movement and conditions like climate, road condition, natural calamity etc. And ensues presenting consistency alongside a trajectory. We combine these deep learning algorithms to get accurate and perfected value for development of future intersections and improvising the currents intersections. The proposed model enhances the TM counts' accuracy and helps in efficient traffic analysis. These TM count can be used for many other purposes in the future like traffic forecast and analyzing the traffic movements.