I develop biologically-inspired algorithms for optical flow, stereo, and camera motion estimation, and combine them into real-time systems that provide robust task-relevant information. Examples of such information are moving object detection and 3D object tracking, which are important in navigation and object manipulation tasks.
To be useful in real-world scenarios, this information needs to be extracted robustly. The methods I develop are insensitive to illumination changes, image noise, and (unstable) camera motion. They are also invariant to certain scene and object characteristics, such as 3D shape, appearance, and rigidity.
I rely on the interplay between computer graphics and computer vision to achieve real-time performance. Taking inspiration from biological vision allows me to more easily exploit the parallelism provided by Graphics Processing Units.
Most of the source code is available as part of the SimTrack package for real-time model-based object pose estimation. Have a look at my YouTube-channel for some recent results. Here are some examples:
Real-time pose estimation of hundreds of objects
Real-time object pose estimation under imprecise calibration
Real-time articulated object pose detection and tracking