Augmented-reality and mirrorworld

Transcription

1 Transit and Transport A Lightweight Platform for Web Mashups in Immersive Mirror Worlds Cloud City Scene is a lightweight platform that enables visualizations of Web mashups in an immersive mirror-world environment in which annotations blend in with buildings, terrain, and objects, letting users interact with the underlying real-world scene. Vlad Stirbu, Yu You, Kimmo Roimela, and Ville-Veikko Mattila Nokia Research Center Augmented-reality and mirrorworld applications are a new breed of applications that belong to the mixed-reality section of Paul Milgram s virtuality continuum. 1 Although augmented-reality applications can overlay digital artifacts using a see-through display for in situ exploration, the mirror-world applications aim to create a realistic, information-enhanced virtual replica of the world, 2 enabling remote exploration scenarios. The objects of the mirror worlds are 3D models automatically constructed from digital pictures, videos, or laser or sensory inputs, acquired via rapid drive-thru (that is, map providers drive a specialized car that lets them acquire images while driving). Exposed in augmented imagery, the 3D models enable an immersive experience that lets users engage with the environment in ways not possible in the real world. 3 Although mirror-world applications provide a rich immersive experience and flexibility in a single application, they don t interoperate well with external applications and programming environments. This limits their appeal, because simply extending the default functionality, or reusing parts of the functionality in other contexts, involves learning specific technologies, leading to a fragmented ecosystem. A Web browser, however, provides a generic runtime environment suitable for a wide range of devices from advanced desktop computers to mobile devices that have already proven successful for visualizing geo-tagged information on 2D maps. Attempts to bring mirror-world immersive experiences into the browser have thus far been limited to the latest browser engines running on high-end desktop computers equipped with hardware 3D acceleration. Although mobile devices are becoming more sophisticated, it will still be some time before they reach the capabilities of their desktop counterparts. With this in mind, we developed Cloud City Scene, our approach to enabling mirrorworld immersive experiences in mainstream mobile and desktop Web browsers. We rely on Web technologies that have been used traditionally for visualizing geo-tagged information on a 2D canvas to achieve an immersive experience in a 3D-like scene. The component is fully integrated in the browser runtime environment, letting Web developers create mashups that can be visualized inside the immersive scene. The client-side system is complemented by a server component that converts geo-tagged data to a format that can be visualized in the scene that the Web browser displays. 34 PERVASIVE computing Published by the IEEE CS n /13/$ IEEE

2 Related Work in Web Mashups Web mashups emerged as a simple way to create more sophisticated applications by combining, in a homogenous way, data from multiple existing sources. 1 In its simplest form, a mashup aggregates or summarizes different sets of data, or creates alternative user interfaces for a website for situations that weren t envisioned by the designers or that don t fit the primary usage scenario. More elaborate approaches personalize the content to match the specific needs or preferences of each user, or add visualization to raw data, typically overlaid on a map. A relatively new category of mashups focuses on real-time monitoring, the design goal being to make the users aware that the used datasets change. The nature of the change might vary with the update frequencies, varying from days to seconds. 2 In line with the client-server architecture of the Web, the mashups can be implemented either in the browser, using various Java- Script libraries for data processing, or on the server. In this context, mapping mashups belong to the mashups that add visualization to geographically tagged data by displaying them over a map canvas using pins, custom icons, or overlays that contain various geometric shapes. Their popularity, fueled by services such as Google Maps ( com) and Open Street Map ( is captured by the mashup dashboard at ProgrammableWeb (www. programmableweb.com/mashups), where, as of March 2012, close to one in three of all mashups registered on the website is a mapping mashup. Other mashup categories, such as the ones related to photography, travel, or transportation, might also use maps for visualization, which takes the mapping mashups percentage even higher. Tools that enable the 3D visualization of the world take the mapping mashup to a new level. For example, Google Earth ( allows remote exploration of locations on Earth, the Moon, and even Mars, while the geographically tagged data can be mashed up using Keyhole Markup Language (KML) documents, 3 or Collada documents 4 for imported synthetic 3D models. Similarly, native augmented-reality applications developed by academia or industry, such as Argon Browser ( or Layar Reality Browser ( allow in situ exploration of augmented reality mashups that are presented to users as Kharma channels, 5 a KML extension that lets Web content be geographically positioned and dynamically manipulated using JavaScript or layar visions. Web mashups have democratized what used to be the domain of custom-built Geographic Information Systems (GIS). Now, even small developers can take geo-tagged data exposed by a plethora of Web services and visualize them on a map. The trend is accelerating as central and local governments expose data of public interest as Open Linked Data ( ReferenCES 1. J. Yu et al., Understanding Mashup Development, IEEE Internet Computing, vol. 12, no. 5, 2008, pp j. Wong and J. Hong, What Do We Mashup When We Make Mashups? Proc. 4th Int l Workshop on End-User Software Engineering (WEUSE 08), ACM, 2008, pp T. Wilson, OGC Keyhole Markup Language, 2.2.0, Open GIS Consortium, m. Barnes and E.L. Finch, COLLADA 3D Asset Exchange Schema, Release 1.5.0, Khronos Group, A. Hill et al., Kharma: An Open KML/HTML Architecture for Mobile Augmented Reality Applications, Proc. IEEE Int l Symp. Mixed and Augmented Reality (ISmar 10), IEEE, 2010, pp Web and Resource-Oriented Mashup Environment Here, we provide an overview of the system architecture, describe the browserbased rendering library and scripting engine that let application developers visualize Web mashups in augmented and mirror worlds, and present the Web infrastructure that provides the data that the rendering engine uses. (For more general information, see the Related Work in Web Mashups sidebar). System Architecture The Cloud City Scene system architecture expands the reach of augmented- and mixed-reality applications beyond the established core of dedicated native applications. The architecture, depicted in Figure 1, has two major components: the server back end, responsible for processing the geo-tagged data and exposing the result as Web resources, and the Web applications running on the user s devices, responsible for rendering the information and handling the user interaction. The Cloud City Scene server takes as input the information provided by services that expose geographically tagged data consumed in native applications, such as panorama textures, terrain meshes, and building models. It then converts them into formats that are appropriate for use in the mainstream desktop and mobile Web browsers. The Cloud City Scene client is a Java Script library that interacts with the server and displays the information to the user. The client can handle two classes of Web browsers. The first group includes the full-fledged desktop and mobile browsers that support a range of HTML5 technologies and have their own JavaScript engine. The second category includes hybrid Web browsers (such as the Nokia Xpress Browser) targeted at resource-constrained mobile january march 2013 PERVASIVE computing 35

3 Transit and Transport Geo-data Server back end User devices Technologies Panorama textures Building models Terrain meshes Virtual artifacts Geo-data server Augmented/mixed reality application Native applications WebGL browsers Panorama projections Building and terrain masks Points-of-interest (POI) placement masks Virtual artifacts projections Cloud City Scene server Cloud City Scene client library HTML5 2D canvas context JavaScript Cloud City Scene domain Proxy browser server (JavaScript engine) Proxy browser client Image map CSS2.1 (JavaScript) Figure 1. The Cloud City Scene system architecture. It has two major components: the server back end and the Web applications running on the user s devices. devices in which the JavaScript processing is done in a browser proxy hosted in the network while the optimized results are sent to the mobile device. The approach lets a wide range of devices consume content that s visualized in realistic 3D-like fashion. Because Cloud City Scene relies on Web technologies, it can target multiple platforms, and the application developers can maintain consistency without worrying about device particularities. The Viewer Cloud City Scene is a mirror-world application that lets users interact with a realistic 3D replica of the real world. The application relies on a set of HTML5 technologies that are readily supported by the mainstream desktop and mobile Web browsers, allowing its use on a wide range of mobile devices, desktop, and laptop computers, and even consumer electronics devices. The Mirror-World Scene. The Cloud City Scene viewer application uses mirrorworld scenes to achieve the immersive viewing experience. Each scene contains information about objects that are visible in the physical proximity of a geographic location that corresponds to a street-level panorama image viewpoint. The objects visible in one scene are represented as polygon masks that are projected against the background panorama image. In addition to the mask, an object can have additional metadata, such as the distance from the viewer and the angle under which the object is seen by the viewer (see Figure 2a). Furthermore, the masks let users interact with each corresponding object in a customizable fashion. Although the scene is viewed by the human user as a 2D canvas, 4 the Cloud City Scene application maintains the information about the objects in a scene graph data structure. This lets the viewer recreate the perception of depth using a z-index, which displays the masks as layers ordered according to their distance from the viewer (see Figure 2b). The panorama image is projected onto the outer layer, followed by the building and terrain masks. The scene can also contain virtual objects that don t exist in the real world. We can group the objects in the scene, based on their nature, using the following categories. Building masks represent the building visible from the location where the panorama image was taken. Each mask has an associated building unique identifier, which lets the users interact with the corresponding building. Terrain masks represent the streets visible from the panorama location. The terrain masks let the users navigate from one panorama location to another. Points-of-interest (POIs) placement masks correspond to positions on building facades on which information about the POIs associated with the respective building can be placed. Each mask contains metadata about the distance from the center of the panorama and viewing angle. The information is used to shrink or tilt the POI representations to create the perception of depth. The placement masks let the mashup application easily place content with which the user can interact. Virtual artifacts represent synthetic objects that don t exist in the real world. A virtual artifact binds together a 36 PERVASIVE computing

4 Virtual artifact Building Heading Projected panorama Far Building and terrain masks POI placement or object mask View angle POI placement or object mask Close (a) Building Mask Viewer Scene Depth Building Visible world (b) z-index (c) Figure 2. A mirror-world scene: (a) the metadata elements, (b) the scene structure, and (c) a scene example, with building masks (purple), terrain masks (red), and points-of-interest (POI) placement masks (blue), near Portsmouth Square in San Francisco. geographic location, a 3D model, and Web content, letting the object be viewed in the mirror world. 5 The virtual artifact masks let users interact with the corresponding virtual object. The Scripting Engine. Cloud City Scene lets application developers interact programmatically with the rendered scene using a JavaScript API. The API provides a set of utilities that enable scene movement, awareness, and alterations. The movement functionality lets the developer change the scene to a different location or adjust the viewport heading inside a scene. The awareness functionality provides feedback about the objects present in the scene or visible in the field of view. The alteration functionality lets developers insert virtual artifacts into the scene. To import a virtual artifact, the developer provides the virtual artifact URI, which the back-end server fetches and converts into an object mask rendered into the scene. The library and the back-end server handle the scene placement automatically, taking into account the distance from the viewer and possible occlusions by other objects in the scene. Additionally, developers can register callback functions on all scene objects, which lets them customize the user s interaction with the scene. The Back-End Infrastructure The scene metamodel that Cloud City Scene uses is based on the same urban model data as our native Nokia City Scene client ( com/apps/nokia-city-scene). In Cloud City Scene, this same data is exposed via Web APIs that enable third-party mashup applications to embed their proprietary data into real-world scenes. In general, there are two sources of geographically tagged data: either from general-purpose providers, such as GeoNames, SimpleGeo, and Linked- GeoData, or from domain-specific information or content providers, such as local city public services. Most of this data is already available programmatically, so overlaying geo-data on top of the 2D map is possible. However, the limitations that come with presenting information on a 2D map have motivated research into other presentations, such as Cloud City Scene. The back end is actually a content process pipeline that adds a new dimension to the existing 2D map-based location-aware applications. january march 2013 PERVASIVE computing 37

5 Transit and Transport Geo-tagged data Registration To mash up existing Web content, the application developer might dynamically import an artifact into the scene using the client JavaScript API. Or, he or she might register the data to the back end in advance by publishing either the data in the Keyhole Markup Language (KML) format or a URI back-link to the provider s Web API. The latter approach adds an additional cost for real-time handshakes but offers more dynamic features. The pipeline parses the input data, no matter how it s registered, and translates it into an internal 3D scene graph for server-side rendering (see Figure 3). The scene graph can be an aggregation of multiple input data sources. The rendering module takes the output of the aggregation module and produces an annotated 2D image of the scene as seen from the current viewpoint. Figure 2c exemplifies the output of the rendering module. In addition to the projected model, a reference to the original data source can be passed through for the third-party content items. This way, the mashup client can render application-specific information in a proprietary fashion. Scene Metadata and Server-Side Projection. Cloud City Scene employs three basic layers of information as the background for building mirror-world applications: panoramic imagery, 3D building meshes, and a 3D terrain mesh. Navteq ( captures the data using special data collection vehicles and processes and aligns it into a coherent world model. 6 However, the processing logic in Cloud City Scene is generic and applicable to other data resources and formats. Aggregated 3D scene graph Server-side rendering 2D annotated image Figure 3. Back-end pipeline for the content projection process. The scene graph can be an aggregation of multiple input data sources. To convert the 3D scene into an interactive 2D canvas for the client, we first select the panoramic image closest to the desired viewpoint. This image is reprojected into a 360-degree spherical image, normalized so that the X coordinate of the image directly maps to world-space direction for example, North is always at the center of the image. The 3D scene geometry visible from the chosen viewpoint is then projected into the same 2D Cartesian coordinate system. The visible regions of each building are then segmented into 2D outlines and simplified to polygons with a fairly low number of points each for quick processing in the client. The street network information is used to generate the terrain masks. Size culling is also applied, whereby areas too small for meaningful interaction are dropped altogether. In addition to the 2D polygons corresponding to visible building and terrain regions, we generate 3D metadata for layering additional content on top of the scene and navigating between different views. For this, we further segment the 3D building masks based on 3D surface orientation information so that, effectively, every individual facade of each building becomes a separate patch of the whole building mask. Each patch stores information about its (average) distance from the current viewpoint and angle between the patch s normal and the viewing direction. Aligning Content and Panoramas. Our basic data assets are accurately aligned to world coordinates, but this isn t necessarily true of content we would like to display on top of the basic scene. When content is registered by third-party services with only World Geodetic System (WGS) 84 coordinates (latitude, longitude, and optional altitude), our alignment process searches for the best placement in the 3D coordinate in relation to the nearest panorama. If more data, such as buildings, is available, the process can also try to place the content close to the sides of the streets, facing the same direction as the closest building if possible. This process infers the building model closest to the content coordinate by computing the distance to the center of the mass of each surrounding model. The process then identifies the relevant facade for 3D by casting a ray toward the closest model and observing which facade is intersected. The process can also infer, if needed, the appropriate content altitude from the terrain elevation, allowing the content altitude to be defined relative to the local ground level. Although this process is heuristic, it lets the system place content on facades simply using 2D geolocation. Once the placement is determined, the next step in the pipeline is to generate the masks with proper perspective projection. Case Study: Acme Tours Web Application To gain first-hand experience with our platform, we developed a concept mashup application for Acme Tours, which offers hop-on, hop-off city sightseeing tours. The application is intended to run in mobile Web browsers, presenting information about bus routes and the location of the routes stops and providing personalized schedules for each bus stop that display live information about bus arrivals and departures. The prototype environment resembles the one expected in a real-life deployment (see Figure 4). The user s smartphone interacts with three different websites: the Acme Tours website hosts the Web application and a set of Web resources that correspond to the bus routes and stops, the information signs that display information in the mirror world, and a Web feed that 38 PERVASIVE computing

6 provides live information about the bus arrivals and departures at each bus station; the mapping server provides the Java- Script library and the back-end functionality that lets the routes and bus stops be displayed as an overlay on a 2D map; and the Cloud City Scene service provides the JavaScript library that enables the Cloud City Scene viewer and the corresponding back-end functionality, which lets the information-sign artifacts be imported and visualized in the mirror world. Acme Tours Web application Bus arrivals feed Bus routes and stops Information signs (virtual artifacts) Web browser api.maps.nokia.com Nokia Maps JavaScript API cloudcityscene.net Viewer JavaScript API Back end We hosted the Acme Tours and Cloud City Scene services used in the prototype environment, and we used Nokia Maps, a commercially available solution, as the mapping service ( api.maps.nokia.com/2.0.0/devguide/ overview.html). The Acme Tours application includes a Web-based component that takes the information about the bus routes and stops, exposed as resources by the Acme server, and visualizes them on the 2D map. The application also has a server-based component that takes the information sign exposed by the Acme server so that the Cloud City Scene server can then convert the information into a layer suitable for display in the mirror world viewer. The representations exposed by the resources hosted by the Acme Tours server use well-known, Web-friendly formats. For example, the bus routes and stops are KML documents that can be visualized, without further processing, and overlaid on the map canvas provided by Nokia Maps. Similarly, the information signs are virtual artifacts, a bundle of KML and Collada documents, that can be imported into the scene through the Cloud City Scene server. Finally, the bus arrival feed is a JavaScript Object Notation (JSON) document, 7 compliant with Live Bus Arrival data format (see syndication), which can be handled directly by JavaScript. Figure 4. The prototype environment. A smartphone interacts with three different websites the Acme Tours website, a mapping server, and Cloud City Scene. (a) Figure 5. Acme Tours mashup application using the Cloud City Scene platform: (a) routes and stops visualization on the map, near Trafalgar Square in London, using Internet Explorer 9, and (b) immersive visualization of an information sign virtual artifact using Safari Mobile. From a user s perspective, the application appears homogeneous, even if the data and user interface components are combined from different sources. The user starts the application by typing the URI into the (b) address bar or reading the URI from a 2D barcode printed on the ticket. The application then presents the map view that displays the bus routes and the stops operated by Acme Tours (see Figure 5a). The device s january march 2013 PERVASIVE computing 39

7 Transit and Transport TABLE 1 A summary of the features of immersive mashups solutions. Cloud City Scene Google Earth Argon Browser Layar Reality Browser Supported devices Desktop computers, mobile devices, Desktop computers Mobile devices Mobile devices and hybrid (proxy) browsers Browser integration Default browser engine Plug-in n/a n/a Device sensors access World Wide Web Consortium W3C device APIs Native Native (W3C) device APIs 3D scene awareness Yes Yes No No Programming environment HTML5 HTML5 Kharma Client software development kit current location is displayed as a blue dot, letting the user find the closest bus stop. By tapping on a bus stop marker, the view changes to immersive mode, letting the user explore the bus stop s location by panning the panorama image (see Figure 5b). Each of the scenes that correspond to the bus locations are augmented with the Acme Tour information signs that display information about arriving and departing buses, targeted for each user. The information signs are interactive, letting the users visualize an expanded view of the bus stop schedule. Concept Discussion and Experience Mashups that bring Web content into immersive mirror worlds can be easily created by application developers who don t have large resources at their disposal (see Table 1). Using the Cloud City Scene platform is as easy as creating a traditional Web mashup that displays geographically tagged data on a map canvas. The JavaScript viewer library takes over the ownership of a <div> element in the application webpage and provides convenient methods for controlling the scene location and viewport heading. Developers can also insert virtual artifacts into the scene and register callbacks that handle the user interaction with the scene objects. The learning curve for integrating the functionality in a Web application is small, because the development follows design patterns that are already established on the Web. Our case study application uses relatively simple data exposed by one service. However, running the mashup in the browser runtime environment lets developers import data from various sources by including in the webpages links to the appropriate JavaScript libraries that process the data. Additionally, the browser runtime environment gives access in a cross-platform fashion to sensor or device context data using standard APIs, defined by the Device APIs working group in the World Wide Web Consortium (W3C). Application developers with access to data from local sensors such as positioning, accelerometer, or gyroscope sensors can use the data to create a personal experience for their mashups. Browser-based applications that rely on Cloud City Scene to display geographically tagged information in an immersive mode using street-level panorama images let application developers control how the information is presented to the users. In contrast, native augmented- and mixed-reality applications lack this flexibility, because they have only limited ways of customizing the user interface. For example, the Argon Browser and Layar Reality Browser let developers import content using channels or layar visions, respectively. However, the formats by which these augmented-reality browsers import content are tightly coupled with the viewers, which end up creating content silos around each browser application. To create content that can be visualized in multiple clients, application developers must transform the data into the appropriate format on the server. Another side effect of the tight format coupling is that although the technologies used are Web friendly, the developers can t make browserbased mashups, preventing them from offloading processing tasks to the clients. To compensate for this limitation, developers must reserve additional computational resources on their server infrastructure. Alternatively, the immersive experience enabled by Google Earth can be embedded in Web browsers using a browser plug-in. This approach has limitations, because functionality availability is restricted only to the platforms and browsers supported by the plug-in provider currently desktop browsers running on Windows and Mac OS. The map view provides a familiar metaphor for visualizing geographically tagged data. Despite recent improvements from major map providers that display landmark buildings using 3D-like wireframes, the map view remains basically a view of the world from above. This visualization modality works best at large scale, but it s limited to the microscale. The street-level visualization brings better results, because it matches what a user would see in the real world and enables a finergrained spatial localization. 40 PERVASIVE computing

8 New solutions from the major map providers (including Google Maps and Nokia Maps) rely on WebGL 8 to create immersive experiences at the city level, using maps augmented with 3D building models, and at the street level. However, these solutions have hardware and software requirements that aren t met currently by mobile devices. Our solution works with technologies used for visualizations on a 2D canvas, but because we have a scene graph that manages the scene s spatial information, we achieve 3D-like awareness. For example, when rendering information about points of interest or inserted objects, we consider the distance from the viewer or possible occlusions. Cloud City Scene demonstrates that lightweight Web technologies, used typically for information visualization on a 2D canvas, can be used effectively to create a realistic and immersive street-level representation of the physical world inside Web browsers. The 3D-like environment is fully integrated with the browser runtime environment, allowing a large pool of application developers to create Web mashups that visualize the Web content in the mirror world. The low system requirements not only provide similar features as the more resource-intensive solutions that are fully 3D aware but also expand the reach of mirror-world Web mashups to a wide range of devices. Moving forward, the key challenges will be to expand the mirror worlds beyond the discrete fixed points, where the panorama images were taken, to a continuous space, where the appropriate imagery is generated on the fly. This will help accommodate both different positions on the surface of the world as well as different elevations. Additionally, the virtual artifacts visualized in the mirror world must be better integrated into the environment, taking into account environmental factors, such as illumination. AcknOWLEDgMEnt We acknowledge Finish Funding Agency for Technology and Innovation (TEKES) for funding the research presented in this article. REFEREnCES 1. P. Milgram and F. Kishino, A Taxonomy of Mixed Reality Visual Displays, IEICE Trans. Information Systems, vol. E77-D, no. 12, 1994, pp J. Smart, J. Cascio, and J. Paffendorf, Metaverse Roadmap Overview, 2007; index.html. 3. C. Dede, Immersive Interfaces for Engagement and Learning, Science, vol. 323, no. 5910, 2009, pp ; www. sciencemag.org/content/323/5910/66. abstract. 4. I. Hickson, HTML Canvas 2D Context, World Wide Web Consortium (W3C) working draft, Mar. 2012; TR/2012/WD-2dcontext the Authors Vlad Stirbu is a mobile Web researcher and practitioner. He completed the work described in this article while he was a senior researcher with the Live Mixed Reality Team at the Nokia Research Center in Tampere, Finland. His research interests include Web-based service architectures, location-based services, and GUI toolkits. Stirbu received his PhD in software systems at the Tampere University of Technology. He s a member of IEEE and the IEEE Computer Society. Contact him at vlad.stirbu@ieee.org. Yu You is a senior researcher in the Live Mixed Reality Team at the Nokia Research Center in Tampere, Finland. His research interests include mobile runtimes, Web technologies, service architectures, and cloud computing. Lately he has been engaged in projects related to augmented- and mixed-reality domains and location-based services. You received his PhD in information science from the University of Jyväskylä. Contact him at yu.you@nokia.com. Kimmo Roimela is a principal researcher of computer graphics and mixed reality at the Nokia Research Center in Tampere, Finland. His research interests include real-time rendering, graphics data compression, augmented reality, context-based applications, urban modeling, and physically based illumination. Roimela has an MSc from the Tampere University of Technology. Contact him at kimmo.roimela@nokia.com. Ville-Veikko Mattila is a senior manager at the Nokia Research Center. His research interests include audio and perception-based signal processing, human perception, and user experience, and his later research has focused on mobile multiplayer gaming and mixed and augmented reality. Mattila received his Dr. Tech. in information technology from Tampere University of Technology. Contact him at ville-veikko.mattila@nokia.com. 5. V. Stirbu, D. Murphy, and Y. You, Open and Decentralized Platform for Visualizing Web Mash-Ups in Augmented and Mirror Worlds, Proc. 21st Int l Conf. Companion on World Wide Web (WWW 12), ACM, 2012, pp T. Pylvänäinen et al., Automatic Alignment and Multi-View Segmentation of Street View Data Using 3D Shape Priors, Proc. 5th Int l Symp. 3D Data Processing, Visualization and Transmission (3DPVT 10), 2010; tu-muenchen.de/3dpvt2010/data/ media/e-proceeding/papers/paper033. pdf. 7. D. Crockford, The Application/JSON Media Type for JavaScript Object Notation (JSON), IETF RFC 4627, July 2006; 8. C. Marrin, WebGL Specification, The Khronos Group, Feb. 2011; org/registry/webgl/specs/1.0/. Selected CS articles and columns are also available for free at january march 2013 PERVASIVE computing 41