Olivier Martinot

Graph mining for object tracking in videos

This paper shows a concrete example of the use of graph mining for tracking objects in videos with moving cameras and without any contextual information on the objects to track. To make the mining algorithm efficient, we benefit from a video representation based on dynamic (evolving through time) planar graphs. We then define a number of constraints to efficiently find our so-called spatio-temporal graph patterns. Those patterns are linked through an occurrences graph to allow us to tackle occlusion or graph features instability problems in the video. Experiments on synthetic and real videos show that our method is effective and allows us to find relevant patterns for our tracking application.

Gesture interactions with video: From algorithms to user evaluation

In the context of immersive communications that aim to enable natural experiences and interactions among people, objects, and the environment, we propose a method to enable natural video interactions through hand gesture recognition between users and a video meeting system. An end-to-end study was performed: we started with the development of specific gesture recognition algorithms and concluded with a user evaluation to validate our results. Gestures and their associated functionalities were identified via a user survey which focused on distinguishing two concepts which are often confused: hand posture and hand gesture (i.e., static versus dynamic). Our recognition process was composed of two main tasks: hand posture recognition (i.e., skin segmentation, background subtraction, region combination, feature extraction, and classification) and hand gesture recognition (tracking and recognition). Our approach combined a signal similarity study with a data-mining tool for dynamic gesture recognition. We focused on the experimentation and user evaluation to improve our approach, taking into account user feedback and analyzing performance in different environments and for different users.

Active TV: Enter into a Cross Media, Community-Based Experience

The Active TV concept offers solutions providing end users with easy access to new audiovisual services and applications from various terminals. A unique experience with new types of group interactions becomes possible in the form of advanced local and community services. To enable the delivery of such services with the required quality and security, Active TV addresses the problem of pushing a part of the service logic as close as possible to the end-user - that is to the residential gateway and end-user equipment.

SlideWorld: A multidisciplinary research project to reinvent the videoconferencing user experience

SlideWorld is a Bell Labs research project that aims to create a new immersive user experience in the situation of videoconferencing. It addresses two types of user experiences: the remote listeners immersion in the event and the speakers augmented feedback channel from the remote listeners. Such innovation may leverage technical novelties (such as smile detection) or usage novelties (such as virtual neighbors selected in a crowd). SlideWorld is an example of a successful project that managed to innovate by smoothly integrating technical and usage novelties. To do so, a multidisciplinary team of computer scientists, ergonomists, and designers ran several rounds of design, development, and experimentation to define/refine the user experience and its actual implementation. This paper exemplifies this typical multidisciplinary work.

WellCom: When Mobile Terminals Meet Television

The Eureka-ITEA2 WellCom project addresses issues related to the creation, delivery and management of advanced multimedia applications and services in the context of distributed home environment. To enable the delivery of such services with the required quality and security, WellCom proposes to push a part of the service logic as close as possible to the end-user. The paper presents some results of the WellCom project enabling to create synergies between the TV screen displaying multimedia content and mobile terminals. A unique experience with new types of group interactions (i.e. advanced local and community services) will be proposed.

Unsupervised Tracking from Clustered Graph Patterns

This paper shows how data mining and in particular graph mining and clustering can help to tackle difficult tracking problems such as tracking possibly multiple objects in a video with a moving camera and without any contextual information on the objects to track. Starting from different segmentations of the video frames (dynamic and non dynamic ones), we extract frequent sub graph patterns to create spatio-temporal patterns that may correspond to interesting objects to track. We then cluster the obtained spatio-temporal patterns to get longer and more robust tracks along the video. We compare our tracking method called TRAP to two state-of-the-art tracking ones and show on four synthetic and real videos that our method is effective in this difficult context.

Cloud-Based Automatic Video Editing Using Keywords

In this paper we propose a cloud-based video editing application designated to manage abundance of video using automatic video editing and summarization algorithms. Our approach uses text processing, statistic analysis and information retrieval techniques to analyze audio transcripts of video corpus. A model approach of mashup based on keywords is proposed. A video editing testbed has been designed and implemented. Finally the video editor engine is deployed on Openstack cloud computing platform which deliver a massively scalable cloud operating system.

Implicit Media Knowledge

These years, the number of visual media (e.g. images, photos, videos…) is exponentially increasing with the apparition of new technologies. Most of these media are shared on the Web with the apparition of social networks such as FaceBook, MySpace, Flickr… The problem related to the media extraction and indexing is become a big issue as the need is become primordial to efficiently retrieve the right media among millions of media. Firstly, we will present an overview of some existing visual media indexing/retrieval systems and its related limitations/improvements. Secondly, we will present a new way to improve media indexing/retrieval without effort: it is based on the analysis of media usage (via implicit user actions on media) in order to enrich media with pertinent information.

Towards an Authoring Tool for Personalizable Multimedia Content

In this paper, we focus on the problems of multimedia content personalization. We propose a new method/tool for authoring personalizable multimedia content. In other terms, our approach aims at helping the author (or the service provider) to generate a personalization process for an multimedia content which will be executed during the delivery of the scene. The elementary operations of this process of adaptation/personalization are implemented by dedicated Web services.