Human Activity Classification Dataset

(used in ICDSC 2009 paper)

Overview

• 12 subjects: subject01, subject02, …

• For each subject, there are two scenarios: Train (training data), Test (testing data)

• 9 actions: Bending, Boxing, HandClapping, HandWaving, Kicking, Running, SitDown, StandUp, Walking

• 12 camera views with labels 100, 101, 102, 200, 201, 202, 300, 301, 302, 400, 401, 402. See following figure for camera spatial layout.

• For each subject and each action, the duration of action was 6 seconds. Image capture rate = 30 fps. Therefore there are 180 images for each <subject, action, camera_view> combination.

• All the 12 camera data were acquired simultaneously. Although no explicit synchronization is used between the cameras, the images are nearly synchronized. Please look at the timestamp.dat file contained in every camera directory for the capture time of every image in that directory.

Figure 1. Camera Spatial Layout.

Acknowledgement

This research was funded by Olympus Corporation, Japan.

How to Obtain Access to the Database

To obtain a username and a password for the database, download the license agreement and send a signed copy of the agreement to Professor Johnny Park (jpark <at> purdue.edu).

Once you have obtained a username and password, click here. If you have any difficulties with accessing the database, please email (rvl-webmaster <at> ecn.purdue.edu).