SHORE™ - Object and Face Recognition
Fraunhofer Institute for Integrated Circuits IIS
- JW Player
SHORE™ - Sophisticated High-speed Object Recognition Engine
SHORE™ is the result of years of research and development activities, which culminated in a highly-optimized software library. This solution enables the detection of objects and faces, as well as extremely fine facial analysis.
The foundation of the versatile SHORE™ solution lies in our extensive experience with detection and analysis technologies and a large database for machine learning. The technology is "trained" by accessing a database of more than 10,000 annotated faces. In combination with the structure-based features and learning algorithms, we can train so-called models that boast extremely high recognition rates:
|Frontal facial recognition||91.5 percent detection rate with a 1 in 10 miss rate||CMU+MIT database|
||94.3 percent detection rate||BioID database|
||6.85 annual mean absolute error||FG-NET database|
Optimizing process-intensive components using a variety of algorithmic and hardware-specific approaches makes it possible to process images in real-time. The following table shows the processing speed for various configurations ( Intel Core 2 Duo 6420 using 1 core, BioID database, 384 x 286 pixels).
|Activated Features||Config. 1||Config. 2||Config. 3||Config. 4|
|Fine eye search
|Analysis of four facial expressions||•|
|Time per image [ms]||9.4||19.3||19.9||22.0|
|Images per second [fps]||107.5||51.8||50.3||45.5|
Please find press information on SHORE™: