Real-Time moving object classification using DPMM for road traffic management in smart cities

Abstract

In a smart city build-up, intelligent transportation system (ITS) is essential for efficient and adaptive control of the transportation system. In order to have a meaningful ITS, availability of statistics such as types of the objects (vehicles, human, etc.) in motion and their statistics and interaction with the environment, are important. Automatic classification of moving objects is an important step for building real-time traffic monitoring systems. This helps to get the statistics of the scene and categorization of the objects into different classes such as human, vehicles of different types etc. It can also give the count of different types of vehicles on the road. Such classification helps to narrow down the search while locating vehicles of interest. We present a new approach for classification of moving objects in videos based on Dirichlet Process Mixture Model (DPMM). The inference scheme has been built using the size of the connected components to identify the objects into meaningful classes. Our experimental validation with publicly available datasets reveals that the proposed method can classify moving objects with reasonably high accuracy applicable to real-time traffic analysis. We also emphasize on how chaotic traffic conditions of the city roads can be, thus highlighting the need for having systematic control of the traffic and rule enforcement.