Target tracking for the automatic control of Pan Tilt Zoom cameras: Capturing close-up video sequences of an object of interest evolving in a large field of view often requires to cover this field of view with tens of cameras. This is especially the case in surveillance and sport coverage contexts. The use of Pan-Tilt-Zoom cameras allows zooming and focusing on an object along its displacement with a single camera, but requires a sufficiently reliable feedback about the target position/trajectory from the image processing module in order to perform high quality automatic tracking.
Automatic Team Sport Coverage: Computer vision tools drive the automatic and democratic production of sport events video reports, by using scene analysis to select an appropriate viewpoint in a static and/or dynamic multi-camera infrastructure.
Ball detection and tracking: Foreground detector reinforcement through trajectory analysis: In the context of team sport events monitoring, the various phases of the game must be delimited and interpreted. In the case of a basketball game, the detection and the tracking of the ball are mandatory. However, one of the difficulties is that the ball is often occluded by players. This project deals with the detection of the ballistic trajectory of a ball thrown between two players or toward the basket. Ballistic trajectories are build on the 3D ball candidates previously detected at each time-stamp from a foreground detector.
Compressive Optical Deflectometric Tomography: The refractive index map allows the optical characterization of complex transparent materials such as optical fibers or intraocular lenses. This research topic addresses the problem of reconstructing the refractive index map of a transparent object from few amount of optical deflectometric measurements. We aim at developing a numerical reconstruction method which makes Optical Deflectometric Tomography compressive and robust to noise.
Foreground object detection: Background modeling and foreground mask extraction are key components of low-level computer vision systems. They aim at extracting moving objects in natural scenes observed with static cameras, and thereby often constitute preliminary steps to object recognition, scene understanding and behavioral analysis.
Fusion-based techniques to enhance images and videos: One of the most successful image fusion strategies is based on the Laplacian pyramid decomposition. In the context of multi-scale fusion, the Laplacian pyramid decomposition has recently been demonstrated to be effective for several challenging tasks to enhance images and videos.
Multiple object tracking with prior detections and graph formalisms: This project considers the tracking of multiple objects within video sequence(s). Fundamentally, it aims at formalising application scenarios in which reliability and the discriminability of the object features vary over time. In order to address problems involving large number of targets, and because automatic detection algorithms have gained maturity, our work assumes that a set of prior and plausible targets detections are available at each time instant.
Adaptive Video Access and Personnalized Summarization: Today’s media consumption evolves towards increased user-centric adaptation of contents, to meet the requirements of users having different expectations in terms of story-telling and heterogeneous constraints in terms of access devices. We propose personnalized summarization mechanisms, and adaptive streaming solutions to address this trend.
Bridging 1-bit and High-Resolution Quantized Compressed Sensing with QIHT: In the framework of Quantized Compressed Sensing, we tried to bridge two extreme cases: 1-bit and high resolution quantization. The requirement of consistency of the reconstructed signal with quantized measurement led us to a new reconstruction algorithm called Quantized IHT (QIHT) that outperforms classical algorithms (IHT and BPDN) at low resolutions.
Video segmentation: Image/video segmentation aims at partitioning the visual frames into non-overlapping areas with different semantical contents. It has tremendous applications in data compression, tracking, augmented reality, activity or object recognition, video annotation and video retrieval. Our group focus on fast and efficient segmentation methods, in such a way to extract content with high-level of abstraction from videos.
Virtual view synthesis: A principal limitation of the conventional video technology is that the viewpoint is fixed by the camera that captures a scene. To overcome this limitation, multiple cameras can be set around the scene, and the set of their two-dimensional images can processed to generate novel images, such as those seen by a (non-existent) "virtual camera".