Vehicles Counting from Video Stream for Automatic Traffic flow Analysis Systems

Abstract

Recently, video-based and real-time vehicle counting become a popular approach for Traffic Flow Analysis (TFA). One of the main objectives of this analysis is to solve the problems that cause traffic congestion including identifying peak hours. Subsequently, this paper proposes an Automatic Video-based Vehicles Counting (AVVC) model for accurate vehicle counting from video streams.The AVCV model includesactive contour which is used to detect whether the object is a vehicle or not, Gaussian distribution which is used for background subtraction and Bilateral Filter which is used for removing shadow and also for smoothing the image. Besides, Kalman Filter is used to reduce the noise in the imagesand Histogram of Oriented Gradient (HOG) and Hough Transform algorithms are used to improve the accuracy of the counting by enabling the model to distinguish between two overlapped objects of vehicles. Hence, our contribution is a strong segmentation algorithm that detects foreground pixels of objects corresponding to moving vehicles.The model is tested and evaluated in terms of counting accuracy and precision using standard dataset video records of three different locations. The AVVC model achieves vehicles counting accuracy of 95.14and precision of92.81% on average.