Files
Abstract
Due to a rapid increase in the amount of available traffic data, several companies and research groups are working on traffic forecasting. Vehicle traffic forecasting is predicting the amount of traffic and the speed of vehicles passing through a point. These data are provided by various sensors such as loop-inductance, microwave radars, laser sensors, and video cameras. Automated analysis on video live feeds in real time contributes to increased sources of data and helps as a redundant system for existing systems in case of unexpected device failures. We present a pipeline for extracting the traffic data from traffic videos which can be capable of filtering the incoming traffic from distractions such as non-vehicles, unwanted camera movements, etc. Calibrating traffic cameras automatically without explicit inputs is another important feature of this pipeline. This proposed pipeline will be capable of observing the count of vehicles passing through a point and their average speed with the help of pre-trained deep neural networks.