Login Form

Editors

Activity recognition results on UCF Sports and Holywood2

Table above shows the results, obtained on UCF Sports dataset (http://crcv.ucf.edu/data/UCF_Sports_Action.php). We report recognition rate with respect to the number...


Read More...

Computational efficiency and parallel implementation

The developed algorithms are computationally effective and the compositional processing pipeline is well-suited for implementation on massively parallel architectures. Many...


Read More...

Motion hierarchy structure

Our model is comprised of three processing stages, as shown in the Figure. The task of the lowest stage (layers...


Read More...

Server crash

After experiencing a total server failure, we are back online. We apologize for the inconvenience - we are still in...


Read More...

L1: motion features

Layer L1 provides an input to the compositional hierarchy. Motion, obtained in L0 is encoded using a small dictionary.


Read More...
01234

Robowood - activity recognition dataset

There exist several relatively demanding data sets for detection and recognition of human activities (for example Holywood2 and UCF Sports). However with these data sets, it is extremely difficult to separate motion from the other information, which is available in the video.

For example, the recordings of golf activities will inevitably contain a lot of green, and diving will contain a lot of blue (UCF Sports). In case of movies (Holywood2), complex environments influence the visual presentation of the activity - for example, driving is often shown from a stereotypical perspective of a passenger and running is shown from the perspective of moving camera. Our motivation was to record video data set of activities under controlled conditions. The data set features humanoid robot HOAP3. We have chosen 18 activities, which are relatively easy recognized by human observers, and roughly correspond to martial arts motion - a compromise between variety of data and abilities of the robot. We recorded the activities from three different viewpoints, and the robot motion includes variable amounts of noise. Background provides zero information about the activity. The new data set, "Robowood", also provides problems of different difficulty (from Robowood2 with 2x90 clips to Robowood8 with 2x600 recordings) and has the same format and interface than Holywood2. Three recordings are presented below.

This website uses cookies to manage authentication, navigation, and other functions. By using our website, you agree that we can place these types of cookies on your device.

View e-Privacy Directive Documents

You have declined cookies. This decision can be reversed.

You have allowed cookies to be placed on your computer. This decision can be reversed.