Proven in practice
It is important for us that our algorithms achieve good results not only in synthetic benchmarks but also in practical applications.
Focus on video processing
Our methods are inherently optimized for video processing, both live and retrospective.
Up to date
Having close contacts with research institutions, we are able to quickly incorporate the latest algorithms into our software.
Deep Neural Networks (DNN) represent the state of the art in image analysis and are the technological foundation of our recognition methods. We train and evaluate neural networks with large amounts of data in order to achieve the optimal combination of recognition performance and speed.
With hardware support (e.g. graphics cards), the high performance of the networks can also be used in real-time applications.
We reliably find the position and location of faces in photos or videos. This usually works despite of unfavorable lighting, rotations, partial occlusion or poor video quality. In video, faces are tracked from frame to frame to form continuous tracks. These tracks can be used for more precise identification than single frames alone.
For facial recognition, so-called templates are extracted, which represent the individual characteristics of each face in a compact way. These templates can then be compared with reference templates of query identities. Creating identities from more than one reference template ("enrollment") leads to better recognition rates than with single template comparison.
Age and gender
In addition to classic identification, the age and gender of individuals can also be estimated by analyzing the face. Thus, you can create statistics on the age and gender distribution of customer groups or start searches for these features ("soft biometrics").