Human Identification at a Distance UMD Database

HID-UMD databases contain walking sequences of 25-50 people in 4 different poses captured by our outdoor Philips G3 EnviroDome camera systems. These databases are used in our gait recognition and face recognition systems for human identification at a distance project. Between February and May 2001 we collected the first dataset containing walking sequences in 4 different poses of 25 people from one of the cameras. In the second dataset, which was acquired during the period of June to July 2001, we expanded the number of subjects into 55 people and used 2 cameras to capture the walking subject simultaneously from orthogonal views. Each image sequence is accompanied by a binary image sequence of segmented foreground as well as a text-formatted ground-truth file describing all the details about data, subjects, and capturing method.

We have recently collected a small dataset of 12 people walking at angles of 0, 15, 30, 45 and 60 degrees to the camera. The idea of this collectection is to test our view synthesis algorithm.

These datasets are available for HumanID participants. To obtain these datasets, we have options as follows:

• DDS4 tapes
• Ecrix VXA-1 tapes
• Direct access through our ftp