Real-Time Computer Vision Group
Sensor data is different. Everyone knows there is noise meaning that a sensor never exactly measures the truth. However, if there was only noise, life would be simple. Instead, the incredibly complex reality creates an incredible number of surprising phenomena that affect sensor data. This is particularly relevant for computer vision where light and shadows, blur and reflections, are only a few examples.
How can a computer interpret sensor data despite these many disturbances? The answer is a combination of intuition and formalism: Judge, which effects are relevant for your application; model these effects in a probabilistic framework; conceive algorithms that find the most likely interpretation of the sensor data.
The world is moving. Hence, a particular challenge is to run a probabilistic algorithm in real-time, because probabilistic methods are often time consuming. Our main focus is on engineering implementations that are both efficient and methodically sound.
News
20.02.2013 CfP: European Conference on Mobile Robots 2013
20.11.2012 CfP: Special Issue on Robot Vision
04.10.2011 Upcoming Event: Spatial Cognition 2012
25.09.2011 Meet us at IROS in San Francisco. Oliver Birbach will present on Estimation and Prediction of Multiple Flying Balls Using Probability Hypothesis Density Filtering, Holger Täubig on Real-Time Swept Volume and Distance Computation for Self Collision Detection and René Wagner on Rapid Development of Manifold-Based Graph Optimization Systems for Multi-Sensor Calibration and SLAM. The conference WiFi is holding up quite well so you can contact us by email.
Topics
Keywords
Simultaneous Localization and Mapping (SLAM), Structure from Motion (SfM), Tracking, (Extended/Unscented) Kalman Filter, Gaussian Mixture Models (GMM), Collision Avoidance, Sports Robotics
