Real Time Video based Vehicle Detection, Counting and Classification System

Abstract

Abstract: City planners have wrestled with traffic challenges for a long time. Better techniques for simplifying the process and analyzing traffic are currently being developed. Both the quantity of vehicles at a specific location over a specific time period and the kind of vehicles can be taken into account for traffic analysis. Such devices have been created for decades, but the bulk of them use sensors to identify the moving cars, such as a couple of proximity sensors to determine the direction of the driving vehicle and to count the number of moving vehicles.

These systems are highly effective and have matured, however they are not very cost-efficient. The problem is that such systems demand routine maintenance and calibration. Creating a vision-based vehicle counting and categorization system is the aim of this project.  In order to do feature extraction and be able to identify and count the vehicles, this system takes still pictures from video. The cars are then categorized by comparing the contour regions to the predetermined values. The comparison of two classification algorithms is the work's significant contribution. Utilizing both the Bag of Features (BoF) approach and contour comparison (CC), classification has been achieved.

Index Terms: Background learning, Foreground extraction, Vehicle classification, YOLO algorithm, COCO dataset.