Find Us On:

Twitter
Facebook
RSS
YouTube
Linked In

 

Join Our Mailing List

 

We send only occasional updates, and we do not share our mailing lists.

Demos & Downloads

Demos:

(Additional demos are available on the Research Highlights page.)

Detecting People Getting Up from Bed
(Pierre Moulin, BingBing Ni)

Falls are the number one cause of preventable injuries in hospitals. Many of these falls occur because patients do not want to "bother" nurses to help them when they get out of bed at night. For fall prevention, it is important to have a highly accurate way to detect when patients are getting out of bed, especially in the dark. Further, the method should not compromise the patient's need for privacy. This demonstration showcases ADSC's new approach to detecting people getting up from bed, which relies on depth plus color cameras. The key technique behind the approach is the modeling of motion features in 3D and fusing of multi-modal features in a learning pursuit, for action representation and recognition. The approach is highly accurate, even in the dark. When used in the depth-only mode, the resulting video stream does not reveal personal details. 

Video (link coming soon!)

 

Identifying Medications from Photos
(Jiangbo Lu, Gang Wang)
Many different pills and capsules look similar. Conversely, different manufacturers make the same drug in different sizes, shapes, and colors. Thus it can be easy to mistake one drug for another. ADSC has developed accurate and efficient approaches for recognizing medications, using ordinary photos or video. This technology can be used to verify that prepackaged single-use packets containing several different medications that a particular patient will take all at once do in fact contain the correct drugs. (Fundamental research challenge: efficient object recognition)

Video (link coming soon!)

Downloads:

ADSC RGBD Activities Benchmark Dataset
(BingBing Ni, Gang Wang, Pierre Moulin)

The availability of low-cost depth cameras has encouraged the computer vision community to investigate depth video as the basis for recognizing actions and activities, with potentially greater accuracy and higher speed than is possible with conventional video. However, to compare proposed new action recognition algorithms, researchers need a reference set of action footage filmed with depth cameras, to serve as a benchmarking database. ADSC's benchmark dataset includes registered depth and color video footage showing people performing activities of daily life. This dataset is freely available to other research groups working in the area.

download instructions