arXiv Preprint

In this repository, you will find a Python code that allowed us to record depth, RGB, and Skeletal data using Kinect v1. You will also find scenarios that were interpreted by a group of computer science students in the auditorium of the University of Sciences and Technologies Houari Boumediene (USTHB), Algeria. In the end, recorded frames can be found in the folders: RGB_images, Depth_data, and Skeleton_data. Depth is saved to JPEG images after normalization. Two Kinect cameras were placed at different positions to guarantee more diversity in the angles of taken images.

• Microsoft Kinect v1
• Windows 10
• Kinect Windows SDK v1.8
• python 2.7
• pykinect
• OpenCV

For each recorded scene there are 3 folders: one for depth frames, the second for RGB frames and the third for skeletons. The dataset is organized as follows:

• For sequences: the folder name contains the number of the camera C, the scene number S, then the take number T <C..S..T..>
• For actions: the folder name contains the same information as sequence except we added the number of the actor P <C..P..A..T..>

This dataset provides 36 action classes including single person actions and two-person interactions. The actions are labeled as shown bellow:

Here are links to download raw data, and the link for selected and annotated images for the task of image captioning:

• Camera 1 Data
• Camera 2 Data
• Selected Images