In this paper, we report findings from a study that compared basic video-conferencing, emergent kinetic video-conferencing techniques, and face-to-face meetings. In our study, remote and co-located participants worked together in groups of three. We show, in agreement with prior literature, the strong adverse impact of being remote on participation-levels. We also show that local and remote participants perceived differently their own contributions and others. Extending prior work, we also show that local participants exhibited significantly more overlapping speech with remote participants who used an embodied proxy, than with remote participants in basic-video conferencing (and at a rate similar to overlapping speech for co-located groups). We also describe differences in how the technologies were used to follow conversation. We discuss how these findings extend our understanding of the promise and potential limitations of embodied video-conferencing solutions.

Our research focuses on improving the effectiveness and usability of driving mobile telepresence robots by increasing the user's sense of immersion during the navigation task. To this end we developed a robot platform that allows immersive navigation using head-tracked stereoscopic video and a HMD. We present the result of an initial user study that compares System Usability Scale (SUS) ratings of a robot teleoperation task using head-tracked stereo vision with a baseline fixed video feed and the effect of a low or high placement of the camera(s). Our results show significantly higher ratings for the fixed video condition and no effect of the camera placement. Future work will focus on examining the reasons for the lower ratings of stereo video and and also exploring further visual navigation interfaces.

Name of an identity is strongly influenced by his/her cultural background such as gender and ethnicity, both vital attributes for user profiling, attribute-based retrieval, etc. Typically, the associations between names and attributes (e.g., people named "Amy" are mostly females) are annotated manually or provided by the census data of governments. We propose to associate a name and its likely demographic attributes by exploiting click-throughs between name queries and images with automatically detected facial attributes. This is the first work attempting to translate an abstract name to demographic attributes in visual-data-driven manner, and it is adaptive to incremental data, more countries and even unseen names (the names out of click-through data) without additional manual labels. In the experiments, the automatic name-attribute associations can help gender inference with competitive accuracy by using manual labeling. It also benefits profiling social media users and keyword-based face image retrieval, especially for contributing 12% relative improvement of accuracy in adapting to unseen names.

In this paper we report on our efforts to define a set of document extensions to Cascading Style Sheets (CSS) that allow for structured timing and synchronization of elements within a Web page. Our work considers the scenario in which the temporal structure can be decoupled from the content of the Web page in a similar way that CSS does with the layout, colors and fonts. Based on the SMIL (Synchronized Multimedia Integration Language) temporal model we propose CSS document extensions and discuss the design and implementation of a proof of concept that realizes our contributions. As HTML5 seems to move away from technologies like Flash and XML (eXtensible Markup Language), we believe our approach provides a flexible declarative solution to specify rich media experiences that is more aligned with current Web practices.

In this paper we discuss communication problems in video-mediated small group discussions. We present results from a study in which ad-hoc groups of five people, with moderator, solved a quiz question-select answer style task over a video-conferencing system. The task was performed under different delay conditions, of up to 2000ms additional one-way delay. Even with a delay up to 2000ms, we could not observe any effect on the achieved quiz scores. In contrast, the subjective satisfaction was severely negatively affected. While we would have suspected a clear conversational breakdown with such a high delay, groups adapted their communication style and thus still managed to solve the task. This is, most groups decided to switch to a more explicit turn-taking scheme. We argue that future video-conferencing systems can provide a better experience if they are aware of the current conversational situation and can provide compensation mechanisms. Thus we provide an overview of what cues are relevant and how they are affected by the video-conferencing system and how recent advancements in computational social science can be leveraged. Further, we provide an analysis of the suitability of normal webcam data for such cue recognition. Based on our observations, we suggest strategies that can be implemented to alleviate the problems.

As commercial, off-the-shelf, services enable people to easily connect with friends and relatives, video-mediated communication is filtering into our daily activities. With the proliferation of broadband and powerful devices, multi-party gatherings are becoming a reality in home environments. With the technical infrastructure in place and has been accepted by a large user base, researchers and system designers are concentrating on understanding and optimizing the Quality of Experience (QoE) for participants. Theoretical foundations for QoE have identified three crucial factors for understanding the impact on the individual’s perception: system, context, and user. While most of the current research tends to focus on the system factors (delay, bandwidth, resolution), in this paper we offer a more complete analysis that takes into consideration context and user factors. In particular, we investigate the influence of delay (constant system factor) in the QoE of multi-party conversations. Regarding the context, we extend the typical one-to-one condition to explore conversations between small groups (up to five people). In terms of user factors, we take into account conversation analysis, turn-taking and role-theory, for better understanding the impact of different user profiles. Our investigation allows us to report a detailed analysis on how delay influences the QoE, concluding that the actual interactivity pattern of each participant in the conversation results on different noticeability thresholds of delays. Such results have a direct impact on how we should design and construct video-communication services for multi-party conversations, where user activity should be considered as a prime adaptation and optimization parameter.

We propose Multi-modal Language Models (MLMs), which adapt latent variable models for text document analysis to modeling co-occurrence relationships in multi-modal data. In this paper, we focus on the application of MLMs to indexing slide and spoken text associated with lecture videos, and subsequently employ a multi-modal probabilistic ranking function for lecture video retrieval. The MLM achieves highly competitive results against well established retrieval methods such as the Vector Space Model and Probabilistic Latent Semantic Analysis. Retrieval performance with MLMs is also shown to improve with the quality of the available extracted spoken text.

We present a method for profiling businesses at specific locations that is based on mining information from social media. The method matches geo-tagged tweets from Twitter against venues from Foursquare to identify the specific business mentioned in a tweet. By linking geo-coordinates to places, the tweets associated with a business, such as a store, can then be used to profile that business. We used a sentiment estimator developed for tweets to create sentiment profiles of the stores in a chain, computing the average sentiment of tweets associated with each store. We present the results as heatmaps which show how sentiment differs across stores in the same chain and how some chains have more positive sentiment than other chains. We also created profiles of social group size for businesses and show sample heatmaps illustrating how the size of a social group can vary.

In this chapter, we discuss the problem of computational inference of aesthetics and emotions from images. We draw inspiration from diverse disciplines such as philosophy, photography, art, and psychology to define and understand the key concepts of aesthetics and emotions. We introduce the primary computational problems that the research community has been striving to solve and the computational framework required for solving them. We also describe datasets available for performing assessment and outline several real-world applications where research in this domain can be employed. This chapter discusses the contributions of a significant number of research articles that have attempted to solve problems in aesthetics and emotion inference in the last several years. We conclude the chapter with directions for future research. Here’s a link to the book.
http://mitpress.mit.edu/books/scene-vision

Video Text Retouch is a technique for retouching textual content found in many online videos such as screencasts, recorded presentations and many online e-learning videos. Viewed through our special, HTML5-based player, users can edit in real-time the textual content of the video frames, such as correcting typos or inserting new words between existing characters. Edits are overlaid and tracked at the desired position for as long as the original video content remains similar. We describe the interaction techniques, image processing algorithms and give implementation details of the system.

It is now possible to develop head-mounted devices (HMDs) that allow for ego-centric sensing of mid-air gestural input. Therefore, we explore the use of HMD-based gestural input techniques in smart space environments. We developed a usage scenario to evaluate HMD-based gestural interactions and conducted a user study to elicit qualitative feedback on several HMD-based gestural input techniques. Our results show that for the proposed scenario, mid-air hand gestures are preferred to head gestures for input and rated more favorably compared to non-gestural input techniques available on existing HMDs. Informed by these study results, we developed a prototype HMD system that supports gestural interactions as proposed in our scenario. We conducted a second user study to quantitatively evaluate our prototype comparing several gestural and non-gestural input techniques. The results of this study show no clear advantage or disadvantage of gestural inputs vs.~non-gestural input techniques on HMDs. We did find that voice control as (sole) input modality performed worst compared to the other input techniques we evaluated. Lastly, we present two further applications implemented with our system, demonstrating 3D scene viewing and ambient light control. We conclude by briefly discussing the implications of ego-centric vs.~exo-centric tracking for interaction in smart spaces.

Telepresence systems usually lack mobility. Polly, a wearable telepresence device, allows users to explore remote locations or experience events remotely by means of a person that serves as a mobile "guide". We built a series of hardware prototypes and our current, most promising embodiment consists of a smartphone mounted on a stabilized gimbal that is wearable. The gimbal enables remote control of the viewing angle as well as providing active image stabilization while the guide is walking. We present qualitative findings from a series of 8 field tests using either Polly or only a mobile phone. We found that guides felt more physical comfort when using Polly vs. a phone and that Polly was accepted by other persons at the remote location. Remote participants appreciated the stabilized video and ability to control camera view. Connection and bandwidth issues appear to be the most challenging issues for Polly-like systems.

Secure authentication with devices or services that store sensitive and personal information is highly important. However, traditional password and pin-based authentication methods compromise between the level of security and user experience. AirAuth is a biometric authentication technique that uses in-air gesture input to authenticate users. We evaluated our technique on a predefined (simple) gesture set and our classifier achieved an average accuracy of 96.6% in an equal error rate (EER-)based study. We obtained an accuracy of 100% when exclusively using personal (complex) user gestures. In a further user study, we found that AirAuth is highly resilient to video-based shoulder surfing attacks, with a mea- sured false acceptance rate of just 2.2%. Furthermore, a longitudinal study demonstrates AirAuth’s repeatability and accuracy over time. AirAuth is relatively simple, robust and requires only a low amount of computational power and is hence deployable on embedded or mobile hardware. Un- like traditional authentication methods, our system’s security is positively aligned with user-rated pleasure and excitement levels. In addition, AirAuth attained acceptability ratings in personal, office, and public spaces that are comparable to an existing stroke-based on-screen authentication technique. Based on the results presented in this paper, we believe that AirAuth shows great promise as a novel, secure, ubiquitous, and highly usable authentication method.

Delay has been found as one of the most crucial factors determining the Quality of Experience (QoE) in synchronous video-mediated communication. The effect has been extensively studied for dyadic conversations and recently the study of small group communications has become the focus of the research community. Contrary to dyads, in which the delay is symmetrically perceived, this is not the case for groups. Due to the heterogeneous structure of the internet asymmetric delays between participants are likely to occur.

Distributed teams must co-ordinate a variety of tasks. To do so they need to be able to create, share, and annotate documents as well as discuss plans and goals. Many workflow tools support document sharing, while other tools support videoconferencing, however there exists little support for connecting the two. In this work we describe a system that allows users to share and markup content during web meetings. This shared content can provide important conversational props within the context of a meeting; it can also help users review archived meetings. Users can also extract shared content from meetings directly into other workflow tools.

Polly is an inexpensive, portable telepresence device based on the metaphor of a parrot riding a guide's shoulder and acting as proxy for remote participants. Although remote users may be anyone with a desire for `tele-visits', we focus on limited mobility users. We present a series of prototypes and field tests that informed design iterations. Our current implementations utilize a smartphone on a stabilized, remotely controlled gimbal that can be hand held, placed on perches or carried by wearable frame. We describe findings from trials at campus, museum and faire tours with remote users, including quadriplegics. We found guides were more comfortable using Polly than a phone and that Polly was accepted by other people. Remote participants appreciated stabilized video and having control of the camera. One challenge is negotiation of movement and view control. Our tests suggests Polly is an effective alternative to telepresence robots, phones or fixed cameras.

In recent years, there has been an explosion of social and collaborative applications that leverage location to provide users novel and engaging experiences. Current location technologies work well outdoors but fare poorly indoors. In this paper we present LoCo, a new framework that can provide highly accurate room-level location using a supervised classification scheme. We provide experiments that show this technique is orders of magnitude more efficient than current state-of-the-art Wi- Fi localization techniques. Low classification overhead and computational footprint make classification practical and efficient even on mobile devices. Our framework has also been designed to be easily deployed and lever- aged by developers to help create a new wave of location- driven applications and services.

Media Embedded Target, or MET, is an iconic mark printed in a blank margin of a page that indicates a media link is associated with a nearby region of the page. It guides the user to capture the region and thus retrieve the associated link through visual search within indexed content. The target also serves to separate page regions with media links from other regions of the page. The capture application on the cell phone displays a sight having the same shape as the target near the edge of a camera-view display. The user moves the phone to align the sight with the target printed on the page. Once the system detects correct sight-target alignment, the region in the camera view is captured and sent to the recognition engine which identifies the image and causes the associated media to be displayed on the phone. Since target and sight alignment defines a capture region, this approach saves storage by only indexing visual features in the predefined capture region, rather than indexing the entire page. Target-sight alignment assures that the indexed region is fully captured. We compare the use of MET for guiding capture with two standard methods: one that uses a logo to indicate that media content is available and text to define the capture region and another that explicitly indicates the capture region using a visible boundary mark.

Live 3D reconstruction of a human as a 3D mesh with commodity electronics is becoming a reality. Immersive applications (i.e. cloud gaming, tele-presence) benefit from effective transmission of such content over a bandwidth limited link. In this paper we outline different approaches for compressing live reconstructed mesh geometry based on distributing mesh reconstruction functions between sender and receiver. We evaluate rate-performance-complexity of different configurations. First, we investigate 3D mesh compression methods (i.e. dynamic/static) from MPEG-4. Second, we evaluate the option of using octree based point cloud compression and receiver side surface reconstruction.

In this paper, we describe Gesture Viewport, a projector-camera system that enables finger gesture interactions with media content on any surface. We propose a novel and computationally very efficient finger localization method based on the detection of occlusion patterns inside a virtual sensor grid rendered in a layer on top of a viewport widget. We develop several robust interaction techniques to prevent unintentional gestures to occur, to provide visual feedback to a user, and to minimize the interference of the sensor grid with the media content. We show the effectiveness of the system through three scenarios: viewing photos, navigating Google Maps, and controlling Google Street View.

We examine the use of clustering to identify selfies in a social media user's photos for use in estimating demographic information such as age, gender, and race. Faces are first detected within a user's photos followed by clustering using visual similarity. We define a cluster scoring scheme that uses a combination of within-cluster visual similarity and average face size in a cluster to rank potential selfie-clusters. Finally, we evaluate this ranking approach over a collection of Twitter users and discuss methods that can be used for improving performance in the future.

People often use more than one query when searching for information. They revisit search results to re-find information and build an understanding of their search need through iterative explorations of query formulation. These tasks are not well-supported by search interfaces and web browsers. We designed and built SearchPanel, a Chrome browser extension that helps people manage their ongoing information seeking. This extension combines document and process metadata into an interactive representation of the retrieved documents that can be used for sense-making, navigation, and re-finding documents. In a real-world deployment spanning over two months, results show that SearchPanel appears to have been primarily used for complex information needs, in search sessions with long durations and high numbers of queries. The process metadata features in SearchPanel seem to be of particular importance when working on complex information needs.

Online video is incredibly rich. A 15-minute home improvement YouTube tutorial might include 1500 words of narration, 100 or more significant keyframes showing a visual change from multiple perspectives, several animated objects, references to other examples, a tool list, comments from viewers and a host of other metadata. Furthermore, video accounts for 90% of worldwide Internet traffic. However, it is our observation that video is not widely seen as a full-fledged document; dismissed as a media that, at worst, gilds over substance and, at best, simply augments text-based communications. In this piece, we suggest that negative attitudes toward multimedia documents that include audio and video are largely unfounded and arise mostly because we lack the necessary tools to treat video content as first-order media or to support seamlessly mixing media.

Creating compelling multimedia content is a difficult task. It involves not only the creative process of developing a compelling media-based story, but it also requires significant technical support for content editing, management and distribution. This has been true for printed, audio and visual presentations for centuries. It is certainly true for broadcast media such as radio and television. The talk will survey several approaches to describe and manage media interactions. We will focus on the temporal modeling of context-sensitive personalized interactions of complex collections of independent media objects. Using the concepts of ‘togetherness’ being employed in the EU’s FP-7 project TA2: Together Anywhere, Together Anytime, we will follow the process of media capture, profiling, composition, sharing and end-user manipulation. We will consider the promise of using automated tools and contrast this with the reality of letting real users manipulation presentation semantics in real time. The talk will not present a closed form solution, but will present a series of topics and problems that can stimulate the development of a new generation of systems to stimulate social media interaction.

3D Tele-immersion enables participants in remote locations to share, in real-time, an activity. It offers users interactive and immersive experiences, but it challenges current media streaming solutions. Work in the past has mainly focused on the efficient delivery of image-based 3D videos and on realistic rendering and reconstruction of geometry-based 3D objects. The contribution of this paper is a real-time streaming component for 3D Tele-Immersion with dynamic reconstructed geometry. This component includes both a novel fast compression method and a rateless packet protection scheme specifically designed towards the requirements imposed by real time transmission of live-reconstructed mesh geometry. Tests on a large dataset show an encoding speed-up upto 10 times at comparable compression ratio and quality, when compared to the high-end MPEG-4 SC3DMC mesh encoders. The implemented rateless code ensures complete packet loss protection of the triangle mesh object and a delivery delay within interactive bounds. Contrary to most linear fountain codes, the designed codec enables real time progressive decoding allowing partial decoding each time a packet is received. This approach is compared to transmission over TCP in packet loss rates and latencies, typical in managed WAN and MAN networks, and heavily outperforms it in terms of end-to-end delay. The streaming component has been integrated into a larger 3D Tele-Immersive environment that includes state of the art 3D reconstruction and rendering modules. This resulted in a prototype that can capture, compress transmit and render triangle mesh geometry in real-time in realistic internet conditions as shown in experiments. Compared to alternative methods, lower interactive end-to-end delay and frame rates over 3 times higher are achieved.