Computer Vision

Computer Vision is the science and technology of analyzing images in an (semi-)automated way in order to improve environment awareness. It typically involves the automatic deduction of the structure and properties of a possibly dynamic 3D world from one or more 2D images of the world. Over the last 30 years the field of computer vision has evolved into a broad mature field which is encompassing many subdomains like scene reconstruction, tracking, object recognition, activity detection, etc.

Research at the LME: Color is a big cue in understanding images and is used in a variety of computer vision tasks ranging from quality inspection to face detection. However, the appearance of an object's color is affected by different factors such as illumination, scene geometry, and scene materials. In reflectance analysis, information on these influencing factors is extracted from images to effectively employ color information. In our research we focus on illumination color estimation and skin reflectance modeling. Multispectral Imaging is a helpful tool in reflectance analysis. The acquisition and analysis of higher spectral resolution images allows us to observe details that are imperceptible to the human eye and to better separate the reflectance factors. High spectral resolution means that for each pixel the visible light spectrum is divided into many narrow bands (typically between 30 and 200). We develop novel tools for interactive visualization and analysis. Motion Estimation and Tracking describes the task of following moving objects in a video sequence. We investigate particular approaches, including optical flow methods, nonrigid registration and particle filters. Two applications we tackle are cloud movement in sky images and autonomous navigation of aerial vehicles such as quadrocopters.

Further Projects:

  • Image Forensics
  • Historical Document Analysis
  • Surgical Video Guidance
  • Emotion Recognition

Snapshot of an interactive multispectral image analysis session.