Evaluation of feature points descriptors' performance for visual finger printing localization of smartphones

Visual information are among the useful sources that can be used to localize a smartphone device. Visual information can be used in the process of image matching based on a pre-built database and Visual Map (A map tagged with images). There are different number of feature detection/description algor...

Full description

Saved in:
Bibliographic Details
Published in:2018 IEEE/ION Position, Location and Navigation Symposium (PLANS) pp. 1002 - 1008
Main Authors: Abadi, I., Moussa, A., El-Sheimy, N.
Format: Conference Proceeding
Language:English
Published: IEEE 01-04-2018
Subjects:
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Visual information are among the useful sources that can be used to localize a smartphone device. Visual information can be used in the process of image matching based on a pre-built database and Visual Map (A map tagged with images). There are different number of feature detection/description algorithms. These algorithms vastly differ in the accuracy, space usage and average matching time for each image. The color information obtained using these methods is typically ignored during the further processing steps. This paper investigates and compares thirteen different descriptors for Visual Finger Printing (VFP) localization of smartphones. These descriptors can be divided into three categories; color histogram, color-moment and Scale Invariant Feature Transformation (SIFT) based algorithms. These descriptors except for the original SIFT are all based on the colored information. Four spaces, RGB, opponent, transformed and HSV are represented in the results. The results show that the SIFT-like algorithms, despite having a high accuracy, takes more time and space that can surpass the real-time implementation requirements on a smartphone. However, the accuracy associated with SIFT algorithms are predominantly higher than moment based, and histogram-based algorithms. Testing also show a slightly better performance for the RGB-SIFT's, Opponent SIFT (O-SIFT) and SIFT, for the investigated experiment setups. However, with different image size, these advantages might vanish or even be reversed. The results also show that with a proper image size reduction and choice of a color space, the practical implementation requirements on a smartphone for SIFT and histogram-based methods can be met.
ISSN:2153-3598
DOI:10.1109/PLANS.2018.8373478