Since the start of my PhD at Imperial College London, I have been focusing my research actvities on human motion capture, with novel applications to both the medical and sports science domains.
I have worked on a wide range of techniques (markerless computer vision, wearable inertial sensors) for a multitude of applications (home healthcare monitoring, sport coaching tools). I am now solely focusing on computer vision-based techniques, which I attempt to enhance with X-ray imaging.
I am currently working with the Morpheo team on markerless human motion tracking  using anatomically-correct skeleton models. I have started by investigating the joint usage of video cameras and X-ray imaging to build accurate human motion models. The complementarity of the two imaging modalities opens novel exciting research directions at the boundary between motion capture and medical imaging. Initial work has already led to some promising results  .
I have worked on the European FP7 RE@CT project (capture, modelling, and 4D rendering), with interests on related subjects such as real-time camera calibration from natural images (sports pitch line detection, lines-points matching), calibration of rolling shutter CMOS cameras  (variations on the bundle adjustment method), real-time free-viewpoint rendering from multiple views , and motion blur reduction (adaptive deconvolution kernel). I have also worked on WebGL-based applications , .
During my PhD and post-doctorate at Imperial College London's Hamlyn Centre, I have worked on various aspects of human motion capture using cameras and wearable inertial sensors. Thus I have worked mainly within two national collaborative projects: SAPHE and ESPRIT.
The SAPHE project aimed at elderly and/or recovering patients at home. My main contributions were to develop methods and algorithms to process data on the camera device itself in order to limit the risks of privacy breach , as well as the fusion of information coming from both cameras and wearable sensors .
The ESPRIT project proposed to use similar capture and processing methods to monitor elite athletes. My main scientific contributions were to implement real-time autonomous player tracking system on an embedded, mobile, and wireless system  as well as a light-weight inertial wheelchair tracking system . I have also worked on a number of other sports such as swimming , rowing , running , climbing , and speed-skating.
I have spent one year developping new tools for immersive visualisation and interaction (stereo display coupled with motion capture) of various types of archeological data (pictures, text, crop marks, ...) overlaid on 3D maps with multiple textures .
Computed Tomography (CT) scanners provide accurate 3D X-ray attenuation models of living organisms, allowing for a range of studies and diagnostics. However, when compared to regular planar radiography, CT scanners are significantly more expensive, expose the patient to higher radiation doses, and require immobility.
In order to tackle these issues, we proposed to combine a regular X-ray imaging device with a set of cameras  . A software suite which estimates a dense 3D attenuation model of a moving sample based on a short capture by a single X-ray imaging device combined with 10 video cameras has been deployed on the novel KINOVIS platform recently installed at the Grenoble University Hospital (CHU).
Dynamic scenes captured simultaneously from multiple cameras can be rendered with the traditional garphics pipeline, i.e. using a single baked texture mapped over a reconstructed mesh. However, significantly more anisotropic details can be rendered by using all originally captured video streams. For this purpose, the original images are fused at run-time based on the currently required viewpoint. In the context of the RE@CT project with the BBC, I have developped a real-time, view-dependent rendering engine from multiple views to replace previous software. The module has been implemented as both a standalone and an OpenSceneGraph library plugin, and defines a file format for multi-view 4D sequences. It also allowed to replace tracked proxy-props (such as fake swords) by virtual models.
In order to allow for live 2D and 3D effects on moving and zooming sports broadcast cameras, an automated and real-time pose and focal length estimation system was developed. For this purpose, the software used natural features such as pitch lines and the calibration core engine described hereafter. The software comprises the following three modules: sports pitch lines detection and labeling, real-time camera pose estimation engine, and GUI for parameter settings and feedback.
In order to avoid the use of large and cumbersome checker calibration charts for multiple camera calibration in large working volumes, a new system based on a lightweight LED wand was designed, developed, and deployed at the BBC. This system was part of a multiple-view capture, 3D reconstruction, and rendering pipeline. It comprises a 12-LED wireless calibration wand, a real-time LED detection and labeling software, and the actual camera parameter estimation. Furthermore, a plugin for the capture module has been developed, which provides real-time feedback during acquisition.
A light-weight sensor composed of an accelerometer and gyroscope was mounted in the wheel of a wheelchair for motion tracking in real-time with wireless feedback to the sports coach. Based on a physical model of the wheelchair intrinsic and global motion, the fusion of several sensors readings with a Kalman filter provides a robust motion estimation, with limited drift over time. The final solution enabled accurate displacement estimation at a limited cost, weight, and installation complexity . The system was trialled by the British Paralympic Basketball team.
While at Imperial College London, I have participated in the development of an autonomous, real-time, wireless tennis player tracker using computer vision  in collaboration with the Lawn Tennis Association (LTA). The hardware was based on a custom-built ARM-like architecture encompassing a camera and a wireless link. I worked on the firmware for autonomous tennis player tracking. It consisted of the following modules: tennis court line detection, camera calibration, background segmentation, estimation of the player on court, and micro web server for direct access.
I have worked on an immersive visualization and interaction of multidimensional archaeological data - Digital Archaeological Virtual Environment (DAVE). The aim was to help archaeologists to classify various kinds of data (text, 3D objects, sketches, images, video, ...) by rendering them onto a 3D terrain model with multiple overlays (aerial photography, maps, ...) .
Various options were included to visualise the scene under different perspectives (eg. rendering an archaeological findings density map, extrapolating the crop marks as walls, changing the sea level, ...). The application rendered on a large back-projected wall screen (6x3m), with a wireless joystick and a tracked glove to enable interaction within the scene.
I am a keen marathon and ultramarathon runner (life-is-an-ultramarathon.org) and also enjoy hiking, rock climbing, paragliding, homebrewing, and photography.
I am particularly involved in the Hardmoors race series in North Yorkshire. Aside from running it, I volunteered as marshal and race director and I have been the hardmoors110.org.uk webmaster for several years.