Detection and Analysis of Objects and Faces
Fraunhofer Institute for Integrated Circuits
Current R&D Activities
Based on existing technologies, we are constantly improving our methods and exploring new research areas. Apart from enhancing our SHORE™ software solution, we are working on new methods for object detection and in-depth analysis of facial expressions.
Face detection and in-depth analysis
Despite their sophistication, our core technologies offer potential for further improvement and expansion. Recent advances incorporated into the latest versions of our SHORE™ software solution include the ability to identify inaccurate detection results (image areas misidentified as faces). In order to improve the accuracy of face analysis, existing methods, for example for age estimation, are being optimized and new ones implemented. In addition, diverse technologies are being optimized for use in sophisticated compact cameras or smartphones.
Object detection
Generic methods make it possible to detect and analyze diverse classes of objects, provided that the objects in a given class have certain structural features in common. One example of how such methods can be successfully applied comes from the ongoing Saisbeco research project, which involves detecting great apes. A crucial element is the ability to distinguish chimpanzees from gorillas. For more information on this subject see "Fast face detection and species classification of African great apes" (Ernst, A., Küblbeck, Ch., International Conference on Advanced Video and Signal-based Surveillance (AVSS) 8, 2011, Klagenfurt).
Experiments involving other types of objects, for instance human hands and road vehicles, have also been conducted. The results are promising and can now serve as a basis for custom adaptations and client-specific solutions.
In-depth analysis of facial expressions through face tracking based on 3D models
Further progress is currently being made with the development of methods for in-depth face analysis based on muscular activity. The aim is to be able to reliably detect and track muscle movements in human faces. The methods under development involve matching faces in images against a parameterizable 3D model. Besides individual facial movements, the position and orientation of faces can also be identified and tracked. These research activities are in connection with the Facial Action Coding System (FACS), a system that was developed in the field of psychology and makes it possible to detect pain, among other things. A technology equipped with this capability would help hospitals use painkillers more effectively, thus improving the wellbeing of seriously ill patients.
Passive 3D location of objects
Many applications benefit greatly from the ability to determine the exact position of faces in space. This is true especially for teleconferencing systems, audio-based speaker location as well as interactive games and exhibits. The technologies developed are passive in the sense that they involve a multi-camera system with no additional active lighting (such as infrared). Several views of a scene are simultaneously analyzed. Based on these analyses, the objects' spatial positions can then be determined. However, precision location necessitates calibrating the camera system. This important step is where object detection intersects with work on methods for automatic camera calibration , another of our research themes in the field of cognitive systems.
Set Bookmark