We collected the Human Activities Under Surveillance – Person Interaction (HAUS-PI) dataset and the Multiple Views version of HAUS-PI, indicated as MVHAUS-PI:
- The dataset comprises videos of 16 person interaction classes with approximately 45 samples per class:
- handshaking (HS)
- hugging (HG)
- highfiving (HF)
- kicking (KI)
- punching (PC)
- pushing (PS)
- slapping (SL)
- bowing (BO)
- waving (WA)
- starring (SR)
- getting up (GU)
- exchanging (CE)
- shooting (SH)
- stabbing (SB)
- talking (TA)
- patting (PA)
- A sketch of the collection site is shown in figure below, where a camcorder, with 1280✕720 of progressive pixel resolution, and three surveillance cameras, with 640✕480 of interlaced pixel resolution, were used to record the activities in the visible area:
- HAUS-PI consists of high resolution videos captured from the camcorder.
- MVHAUS-PI consists of videos captured from PTZ cameras.
- Some image samples of the MVHAUS-PI dataset:
- Some image samples of the HAUS-PI dataset:
More information at: http://vision.csee.wvu.edu/HAUS
Key words: Multi view human interaction recognition, action recognition, multiple camera video surveillance, segmentation, activity and action detection, Cross view action recognition, Multiview benchmark dataset.