Vehicle Detection And Counting From Unmanned Aerial Vehicle (UAV)

Ang, Kuan Kee (2015) Vehicle Detection And Counting From Unmanned Aerial Vehicle (UAV). Project Report. UTeM, Melaka, Malaysia. (Submitted)

Text (24 Pages) Vehicle Detection And Counting From Unmanned Aerial Vehicle (UAV) 24 Pages.pdf - Submitted Version Download (259kB)

Abstract

Efficiency traffic management is important in order to manage the traffic efficiently especially during unexpected situations. To do so, continuous and reliable information on traffic situation is important to users so that they can plan their route by avoiding traffic congestion. UAV images are an important means by which detection and classification of vehicle can be carried out to assist the traffic management. This is done by capturing the latest information of the road situation and reporting it to road users. In this study, I develop a procedure to detect and count the number of vehicles from three different height categories of UAV images. Initially, two unsupervised classification methods, ISODATA and K-Mean, are considered to separate vehicle and non-vehicle pixels from the UAV images. The study were conducted in two phase; phase one is classifying the objects within the image and phase two, determining the total number of vehicles within the image. Firstly, the unsupervised classifications were implemented and analysed to cluster similar pixels within the selected region. Secondly, region resizing is carried out to reduce the portion which is non-relevant from an image. In order to get the total amount of vehicles in an image, a standard amount of pixels for a single vehicle is computed. The accuracy assessment in this study is measured by means of accuracy and error percentage. This is done by comparing between the actual numbers of vehicle with the number of vehicles computed from UAV images. For both classifiers, low category images are found to have the highest accuracies. Further study needs to focus on an automated technique of resizing the region to reduce the non-relevant portion from an image in order to get the exact regions of vehicles.

Actions (login required)

View Item