Vehicle Detection, Classification and Tracking Using Deep Learning

Abstract

Vehicle Detection and Classification is one of the most important tasks in a traffic scenario. In this paper, Deep Learning algorithms utilized for the vehicle detection and classification task. To reduce the road traffic, and to determine the location of the vehicles, vehicle tracking and counting approach are used. The first step for vehicle tracking and counting is vehicle detection. The Hungarian algorithm is used for finding an association between detection to tracker. Once the matches between detection to the tracker is found, the Kalman filter is applied to predict the current state of the vehicle or to correct the state of the matched vehicle. The Mask RCNN deep learning model is used for the vehicle detection task. The vehicle tracking by detection improves the results of vehicle tracking and counting.