The main objective of this proposal is to propose and evaluate a new workflow which mixes user interaction using motion-tracked cameras and automated computation aspects for interactive virtual cinematography that will better support user creativity. We propose a novel cinematographic workflow that features a dynamic collaboration of a creative human filmmaker with an automated virtual camera planner. This process aims at enhancing the filmmaker’s creative potential by enabling very rapid exploration of a wide range of viewpoint suggestions of a 3D environment. It looks at enhancing the quality and utility of the automated planner’s suggestions by adapting and reacting to the creative choices made by the filmmaker. This requires three advances in the field. First, the ability to generate relevant viewpoint suggestions following classical cinematic conventions. The formalization of these conventions in a computationally efficient and expressive model is a challenging task in order to select and propose the user with a relevant subset of viewpoints among millions of possibilities. Second, the ability to analyze data from real movies in order to formalize some elements of cinematographic style and genre. The issue here is to encode these characteristic elements of style and genre and let the users select which genre they prefer on a given scene. In this project we propose to characterize elements of style and genre using reinforcement learning techniques from hand-annotated real movies. Third, the integration of motion-tracked cameras in the workflow. Motion-tracked cameras represent a great potential for cinematographic content creation. However given that tracking spaces are of limited size, there is a need to provide novel interaction metaphors to ease the process of content creation with tracked cameras. Finally we propose an evaluation of our tool involving both professionals and film schools. Proposing tools to interactively assist users in the design of shots, edits and camera paths is a novel and ambitious research path which strongly contrasts with main research in the field, more involved in the design of declarative viewpoint computation, optimization and planning techniques. Furthermore this project displays a strong potential for industrial transfer.

Identity-related frauds represent a major risk on the society safety given its serious consequences. These consequences may vary from small but very frequent frauds (telecom contracts, small credits, etc.) to transnational organized crimes and terrorist actions. An increasing number of false identity documents have been detected during the last few years, according to several official studies around the globe. The fast development of such criminal activities can be explained by an easy and public access to advanced technologies. Several studies have reported the organized nature of ID fraud activities and the progress of such black market. In order to fight against identity-related frauds, traditional investigation methods applied to identity documents (ID) rely on the presence of an expert, which significantly reduces the spread of such important verification in many administrative and commercial entities. In addition, existing ID control tools have shown several limitations, such as high false positive rates (rejection of valid documents), partial controls or nonexistent evolving capacities (new ID models, new control rules). Because of these shortcomings, automatic verification tools have not been widely used and their role has been reduced to data memorizing and simple assistance tasks. When ID frauds are detected, it is important to discover forensic links in order to identify the source of those frauds. Current investigation methods do not sufficiently address this problem and are still based on case-by-case approaches with no global analysis. However, an efficient automatic fraud pro ling system allowing to ID fraud link detection will certainly be of great benefit to anti-fraud authorities and will help to uncover many forgery worldwide networks. The core idea of IDFRAud, our proposition of an industrial research project, is to establish a virtuous circle between two processes: (1) the automatic verification of ID documents, and (2) the automatic profiling of ID frauds. The first process applies control rules on ID documents in order to check their validity, and sends detected ID frauds to the second process that analyzes them in order to discover forensic links (fraud profiling), and to enhance the ID control rules. Control rules are stored and maintained in a knowledge base in order to facilitate the system evolution. The knowledge base is also fed with existing repositories of ID document models (like Prado1) and ID frauds. In fact, adding new control rules enables more robust future ID controls, which in turn enable the detection of more ID frauds, and forensic links. The first originality of IDFRAud is to propose an automatic solution for ID verification that can handle documents issued from a large set of countries. The solution will be able to execute specific controls according to the ID model (type, country, generation, etc) thanks to a knowledge base. ID content and rules modeling is one of the main originalities of IDFRAud. To the best of our knowledge, there is no existing formal description of ID documents and existing public and industrial ID knowledge bases cannot be directly used for automatic reading and verification. ID fraud automatic profiling represents a major ambition of IDFRAud. Experts from national security authorities along with academic and industrial partners will work side by side to propose the first data analysis solution dedicated to ID forensic link detection. Such intelligent solution aims at replacing the manual fastidious analysis that can hardly cope with a high-dimensional evolving false ID datasets.

PhoReVox is a multidisciplinary project whose objective is to provide a tool to help learners develop written skills in French thanks to the use of speech technologies. Technological solutions to assist language learning, and more particularly written French, remain mostly confined to exercises based mainly on the use of written forms (MCQ, keyboarding words, pictograms). We think that an oral interaction can lead to greater control of the written forms of language. a wide range of pedagogical exercises covering different stages of language acquisition will be provided : phonological discrimination tasks, exercises such as dictation of words or phrases, work on punctuation and word order, work on French liaisons and its relation to speech style, etc. All these exercises will in the end seek to help learners develop the skills necessary to produce free written texts. To reach the project objectif, five areas of innovative research will be explored: -Offering voice synthesis of high quality. Accepting and understanding the exercises depend on it. Models of French prosody adapted to different forms of exercises will have to be defined. -Defining types of exercises that can be used to work on specific language skills in different learning modules. -Providing a tool able to vocalize exercises and collect the learner's answers on a kind of Web 2.0 platform. -Monitoring skill profiles. A major challenge consists in automating the collecting process of these indicators. The collected data should be analyzed automatically. -Providing a tool able to create exercises automatically according to the learner's profile. The quality of speech synthesis is crucial. The synthetic voices should keep the learner's attention alive and correspond to what is expected for the proposed activities. Up to now, no speech synthesis system can, for example, say a poem or a nursery rhyme convincingly. The exercises must be tailored to the learners' needs. On this point, managing learners' profiles and generating content automatically are fundamental. This project brings together complementary actors whose aim is to provide within two years a prototype that will be tested both with "cycle 2" primary students and A2-B1 level learners of French as a foreign language. The CREAD, a research centre on education, learning and didactics, will participate with the LLF, a Formal Linguistics Laboratory located at Paris Diderot University, and ZEUGMO company, specialized in learning spelling, to the specification of the contents included in the exercises and ensure the implementation of field experiments. Voxygen company will provide a speech synthesis technology adapted to the objectives of the application, in particularly in terms of new models of prosody associated to dedicated voices. The Cordial research team, which is part of IRISA, will manage the implementation of computational models related to the prosodic specifications defined by LLF as well as take part in the automatic generation of contents for the exercises. An innovative part of the project seeks indeed to reinforce the learner's autonomy. A profile is elaborated based on the results obtained by the learner to various exercises. The learner's profile is then used to provide him with a set of exercises, which can be generated automatically, focused on his main difficulties. The implementation of a demonstration prototype will be integrated into the platform "Orthodidacte.com" provided by ZEUGMO company to help learners acquire spelling skills.