Object tracking and re-identification are challenges common to a number of fields in computer vision research. In camera networks, the usual goal is to build a complete view-to-view track of an object without human intervention. This track can be then used for conventional video surveillance or, for example, to feed another system.
The development of camera network tracking algorithms requires a lot of data, especially when the goal is to learn object models suitable for sequence matching. CMV100 is an indoor dataset designed for such purpose and more. It contains 100 hundred tracked objects appearing in multiple camera views (in five at most). The number of video sequences is over 400. In addition to original video sequences, the CMV dataset contains foreground masks and tracking data (bounding boxes, image descriptors, etc.) in XML format.
To protect people's privacy, the facial images of the persons appearing in the dataset are not allowed to be published in general. If you need to publish examples of faces in the dataset, use images of the object 78 (object_078) or contact the author for other possible image candidates.
Use only for research.