Aerial Based Traffic Tracking and Vehicle Count Detection Using Background Subtraction

Abstract

With the advent of increasing population, the traffic density is also increasing. Thus, it becomes deemed necessary for the urban planners to analyse the traffic condition via video surveillance for a successful improvisation to the existing town planning. The objective of this study was to develop a high altitude video surveillance setup with centroid tracker and compare different variants of background subtraction comprising of K-Nearest Neighbour, Mixture of Gaussian and Geometric Multi-Grid on a vision-based system for road vehicle counting and tracking. This project uses Python as its programming language and Open Computer Vision (OpenCV) as an open-source library for developing a high altitude video surveillance system for vehicle counting and directional motion detection. The designed system was able to achieve high count precision even in difficult scenarios related to occlusions or the presence of shadows. The principle of the system was to install a camera on the pedestrian bridges and track the vehicular traffic congestion by incorporating a unique ID. Moving objects were tracked using different background subtraction algorithm and object tracking was conducted using the centroid tracker. The video processing model was combined with a motion detection procedure, which correctly classified the positioning of moving vehicles depending on the space and time when the experiment was conducted on the site location. From the results it was revealed that both K-Nearest Neighbour and Mixture of Gaussian showed better accuracy with 93% and 100% depending upon the traffic density modalities. Using the proposed setup design, the identification of severe shadows based on solidity can be computed through the nature of the shape and this classification allows its accuracy to be estimated.