Efficient Vehicle Counting by Eliminating Identical Vehicles in UAV aerial videos

Traffic surveillance using Unmanned Aerial Vehicles (UAV's) has gained a lot of attraction in civilian applications and remote sensing tasks. Thanks to its high mobility and large field of view and ability to cover regions at different altitudes UAVs have made a mark in recent years for surveillance. The primary purpose of UAV in traffic surveillance is to monitor the daily activities in the busy traffic areas and report the abnormal activities which may take place. In recent years, many gated campuses such as educational institutions, organizations, shopping malls, etc. have taken steps to keep a track of vehicles trespassing within its vicinity. Vehicle counting is one of the monitoring tasks performed in surveillance to estimate the density of vehicles in an event or areas where traffic congestion is common. In this paper, a vehicle counting framework is proposed to eliminate the problem of redundant vehicle information count when a vehicle has appeared in successive frames of UAV videos. This work demonstrates that the comparison of concatenated three features vectors (Histogram of Oriented Gradients, Local Binary Pattern, and mean RGB value) can be used to recognize identical vehicles in UAV aerial videos.