J4K Java Library
The J4K library is an open source Java library that implements a Java binding for Kinect SDK. It communicates with a native Windows library, which handles the depth, color, infrared, and skeleton streams of Kinect using the Java Native Interface (JNI). This library was developed by Prof. Angelos Barmpoutis. The J4K library is compatible with old Kinect and Kinect 2. Unfortunately the J4K library supports not all features of Kinect 2. We are going to extend this library in the framework of our project.
CAVIAR Test Case Scenarios
This collection includes live video records created in shopping centre in Lisbon and also staged video samples of different cases of people behavior (both usual and abnormal) created at the entrance lobby of the INRIA Labs, Grenoble, France. People walk, meet each other, enter and exit shops, fight, leave packages in the public place, etc. The resolution of video clips is 384x288 pixels. The data is captured at 25 frames per second. Video clips are compressed using MPEG2 and also converted into JPEG file archives. Video records are annotated manually using the CVML language (based on XML). A set of reference points is also provided, that is necessary for calculating physical coordinates of objects in video scenes. This collection was created ten years ago, in 2003-2004 (the EC Funded CAVIAR project / IST 2001 37540), but is still used by researchers all over the world. One can download MPEG and JPEG files coupled with corresponding XML files on the Web Site: https://homepages.inf.ed.ac.uk/rbf/CAVIAR/
BEHAVE Interactions Test Case Scenarios
This collection includes staged outdoor video samples of different cases of people behavior (usual and abnormal). People are walking, running, meeting with others, fighting, etc. These video records are more complicated for analysis than the CAVIAR data set, because they are outdoor ones and contain bicycles and cars. Several video episodes are damaged a bit, because a person was reflected in the window glass accidentally. The resolution of video clips is 640x480 pixels. The data is captured at 25 frames per second. A set of reference points is provided, that is necessary for the calculation of physical coordinates of objects in video scenes. Unfortunately, some coordinates of these reference points are contradictory; please be careful when you calculate the projective transform matrix. Manually created semantic annotations of the video clips are provided in the ViPER format (it is XML-based). This collection was created in 2007 (the UK's Engineering and Physical Science Research Council (EPSRC) project GR/S98146). One can download AVI, JPEG, and XML files on the Web Site: https://homepages.inf.ed.ac.uk/rbf/BEHAVE/ . Note, that usually we divide long BEHAVE video records into separate JPEG file sets corresponding to individual video episodes to be analyzed by logic programs.