Department of Pattern Recognition

This work has been performed in industrial context and presents a whole framework for people detection and tracking in a camera network. The three main processing steps are addressed:
people detection, people tracking in mono-camera context, and people re-identification in multi-camera context. High performances, system autonomy and ease of deployment, and the real-time processing are the most important constraints which have guided this work. Some parts of the proposed work are already integrated and deployed in a commercial product while others are in prototype state and are planned to be integrated in future.

People detection aims to localise and delimits people in video sequences and static images. The proposed people detection is performed using a cascade of classifiers trained using LogitBoost algorithm on region covariance descriptors. A state of the art approach, providing good performances but not applicable for real time is taken as basis and is optimized and improved to process in real time while the detection performances are increased. Our optimization scheme is generalizable to many other kind of detectors based on cascade of classifiers where the whole space of all possible weak classifiers cannot be reasonably tested.

People tracking in mono-camera context aims to provide a set of reliable images of every observed person by each camera, to extract his visual signature for re-identification purpose. It provides also some real world information which are useful to improve re-identification process. It is achieved by tracking SIFT features using a specific particle filter, in addition to a data association framework based on global optimization, which infers object tracking from SIFT points one, and which deals with most of possible cases, especially occlusions.

Finally, people re-identification is performed using an appearance based approach. A state of the art approach, which performs in real time, but provides various performances depending on the input data is improved to provide better performances while keeping the real-time processing advantage. Some of the improvements are specific to this approach while other improvements are more general ones and can be applied to many state of the art approaches to increase their efficiency.