Mobile computing gets the user out of a desk and into the field, devices are getting smaller, cheaper, all while hardware capabilities increase. This presents a host of opportunities, but also a few unique problems.
A few of the problems in the mobile space:
Groups of users with networked mobile devices offer an unprecedented ability to gather, analyze and consume data in real-time -- if data is organized at the time of collection.
sStitch uses several dimensions to help organization and analysis of mobile data.
The core assumption is where a user and their social group are working and what they are collectively doing can be used to compensate for the losses of user interface that accompany mobile devices. This approach is appealing because more users creating more data gives rise to an increasingly useful system, where currently more users and more data makes the system worse. The ‘intelligence’ of the system is distributed across all the small, individual actions of the users, rather than in any centralized analysis.
The current sStitch system has multiple interaction areas -- mobile data organization, laptop data consumption and mobile data consumption.
When a sensor (i.e. images or sound) data is collected on the mobile device the first task is to assign metadata. Some metadata is automatically assigned transparently to the user. This metadata may vary by photo and determined by hardware, such as GPS coordinates, or may vary by the account settings, such as permission level.
Automatically assigned metadata: User ID, time stamp, GPS coordinates, GPS resolution, altitude, permission levels, camera or sensor statistics.
Some of the metadata is chosen by the user -- namely the tag information, but also title and free-text descriptions. To optimize user interaction time we use geospatial databases and social graph data to suggest tags. GPS coordinates are sent over wireless networks in real-time, and the analyzed with respect to how the user and their social group have organized nearby data.
As a user selects geo-socially predicted tags or types them in de novo linguistic methods are used to suggest new tags. Expanding and refining tag suggestions leads to sophisticated and full tag set coverage. The suggestions also evolve over time, as the system ‘learns’ what a user and their social group prefers linguistically.
Automated geotagging with semi-automated linguistic tagging provides a powerful combination -- it becomes easy to sort large photo sets along space, time and tag dimensions.
We have built a web-based interface for the system, which has all the features of an off-the-shelf photo website. There are several features which combined are particularly useful -- open data standards, geosocial search, and geoRSS output.
We value open data standards to make our system flexible and useful, and we’ve fully integrated use the geoRSS standard and KML. At the bottom of every page are links to a number of data types, including geoRSS. For instance you can use our search engine to generate a query by user, tags, time and spatial ranges. That query has an easily accessible RSS feed, and future data meeting the query will be fed to that URL.
The end result is a wide variety of common operating picture and geospatial systems can import our data real-time for further analysis and action. In the caption to the right we’ve brought a map into the CivMil.org website, which is used by humanitarian aid coordination. In the Golden Phoenix exercises coordinated by the US Marines MAG 46 our field data was imported and analyzed by the Swan Island Networks TIES COP as well as analytics software from the i2 Group.
Our approach lends itself to data consumption on mobile devices, through the integration of mapping software, mobile web applications, and SMS technologies. A user can be notified of a query match through SMS, which can then be viewed on native mapping software, and further explored through the phone’s web browser.
The screen shot on the right shows a query for ‘fire’ in the iPhone’s map application. Tapping on the pins takes the user to contextual information, such as driving directions to the location or URL’s for more detailed information.
Usage Example: Organizing a photo of Gammage Auditorium at Arizona State University, a building designed by Frank Lloyd Wright.
The ‘Nearby’ screen takes the users GPS coordinates and retrieves nearby tags, which are then reweighted using information from the users history as well as the history of the users friends. I take a lot of architectural pictures, so the tags relating to architecture float to the top.
If location based suggestions aren’t sufficient the user can type in a tag and get a linguistic expansion. In this case the term ‘arizonastateuniversity’ is added, and suggestions are made incorporating the users historical preferences. In this case architecture is suggested first because of recent architectural choices.
After uploading the picture is viewable on the sStitch website, or through other software packages by importing either geoRSS or KML streams. In the example below the image is viewed in Google Earth.
Copyright © 2008 MTI. All Rights Reserved by sStitch, Onto LLC, Todd Huffman, and Abstract Approach Digital Solutions. Developed by Abstract Approach Digital Solutions in conjunction with Onto LLC.
Version: Ham Solo (1.0.0)