WhereAmI: Energy Efficient Positioning using Partial Textual Signatures

Abstract

Positioning systems can use signatures hidden in a user's environment to identify a location. Images are often used to locate a place by identifying landmarks. In this work, we present the use of texts in an image to identify a user's location. The key intuition behind this work is that a collection of names of business appearing in an image forms a bag-of-words that provides a unique signature for a location. We use Optical Character Recognition (OCR) to detect the texts from an image. However, use of OCR in outdoor settings is resource intensive, and text detection is often error prone in uncontrolled settings. We develop an algorithm that can handle partial errors in the collection of business names to locate the user. Partial errors in text detection are handled by using similarity scores based on approximate text matching. We also limit the resource usage by partitioning the application between the smartphone and a cloud based web service to save energy. We have implemented the positioning system, called WhereAmI, on Android based smartphone. The experimental results show that WhereAmI can be an alternative positioning technique for GPS in terms of accuracy, precision, energy efficiency and positioning latency.