Automated vehicle detection and classification methods

Abstract

Recognition, Localization and Classification of objects, especially vehicles, have been an area of immense research in Computer Vision (CV). Automated Vehicle Detection and Classification on real world traffic and over various terrains has found its use in autonomous cars, intelligent parking systems, automated petrol systems, intelligent vehicle tracking systems, security systems and many more CV based applications. There has been a considerable evolution in the methods adopted before to detect and classify vehicles in real world scenario to those being employed today. Vehicle Detection and Classification (VDC), today, involves prediction of the coordinate-based location of particular category of vehicles in a given input image by means of bounding boxes using Deep Learning and Neural Networks (NNs). Feature calculation carried out by Deep learning-based vehicle detectors are more in use due to its better performance, ease and higher accuracy compared to the earlier employed machine learning techniques involving SVMs and Regression to identify patterns in vehicle frames. Background Detection methods and use of Gaussian Mixture Models for detection and classification are among the oldest attempts made in this problem. This paper reviews some of the crucial methods, by highlighting the different works done so far by researchers across the globe, to process real-time traffic images and videos to eliminate the need for manual surveillance and thus gradually introducing the idea of smart traffic management tools. This paper further points to the history of CV in VDC problem along with the future possibilities.