Wide area video surveillance always requires to extract and integrate information coming from different cameras and views. Re-identification of people captured from different cameras or different views is one of most challenging problems.
In SARC3D project, the authors are studing an approach for people matching with vertices-based 3D human models. People are detected and tracked in each calibrated camera, and their silhouette, appearance, position and orientation are extracted and used to place, scale and orientate a 3D body model. Color features are computed from the 2D appearance images and mapped to the 3D model vertices, generating the 3D model for each tracked person. A distance function between 3D models is defined in order to find matches among models belonging to the same person.
The SarC3D dataset is created In order to test the SARC3D method. It consists of short video clips of 50 people captured with a calibrated camera. To simplify the model to image alignment, they manually selected four frames for each clip corresponding to predefined positions and postures of the people. Thus the annotated data set is composed by four views for each person, 200 snapshots in total.
This dataset is part of the ViSOR project.
D. Baltieri, R. Vezzani, R. Cucchiara, "3D Body Model Construction and Matching for Real Time People Re-Identification" in Proceedings of Eurographics Italian Chapter Conference 2010 (EG-IT 2010), Genova, Italy, Nov. 18-19, 2010