Evaluation of SIFT on Androidbased Mobile Devices and Personal Computers

摘要

The purpose of this paper is to evaluate two algorithms for the feasibility ofobject recognition on mobile devices. The majority of this paper is based uponthe work by Lowe (1999, 2004) with the SIFT algorithm and a discussionregarding the DAISY algorithm by Tola et al. (2010), evaluating the key aspectsof the algorithm and whether it could be a used on mobile devices.The SIFT algorithm has been implemented to run on a desktop machine and anAndroid-based mobile device to test the feasibility of object detection. Theimplementation for the desktop was done for a comparative study between themobile device and the desktop computer. Source-code of the implementation onthe desktop and the Android platform can be found at:https://github.com/iancornelius/SIFT-AndroidThe results from the outcome found that the task of object recognition on mobiledevices is feasible. Further work to be carried out would be the implementationand evaluation of DAISY on the desktop and Android mobile-device to comparewhether DAISY can improve the results gained on single-core devices. Anotheraspect could be how threading would affect the results especially with thecomputation on the extraction of descriptors from images being donesimulataneously.