NB. I now work at the Wageningen University in the Netherlands. Go to: www.gertkootstra.com

Detecting, Segmenting, and Grasping Unknown Objects

My research, in general, deals with the detection, segmetation, and grasping of unknown objects by a robot.

Visual-Based Grasping of Unknown Objects

Grasping unknown objects based on real-world visual input is a challenging problem. In this paper, we present an Early Cognitive Vision system that builds a hierarchical representation based on edge and texture information, which is a sparse but powerful description of the scene. Based on this representation we generate edge-based and surface-based grasps. The results show that the method generates successful grasps, that the edge and surface information are complementary, and that the method can deal with more complex scenes. We furthermore present a benchmark for visual-based grasping.
Read more.

Visual Attention for Object Detection and Segmentation

During my Ph.D. research, I developed a visual attention model based on symmetry. The comparison of the model to eye-tracking data, showed that symmetry is a good predictor of human eye fixations. The model has furthermore been shown to be successful in selecting visual landmarks for robotic localization and mapping.

In my current research, I focus on the use of symmetry for the detection of unknown objects in the environment. I furthermore focus on the problem of bottom-up object segmentation.

Using symmetry to detect unkown objects

In the Gestalt theory, symmetry is considered one of the important cues for figure-ground segregation by humans. Motivated by this, we use symmetry to detect unknown objects in the environment. Based on fixation points selected by the visual attention model, we initialize object segmentation.
Read more.

Fast and Automatic Detection and Segmentation of Unknown Objects

We propose a segmentation method named Automatic Detection And Segmentation (ADAS). Where many existing methods need user input to initiate the segmentation, our method is initiated by our symmetry model. Furthermore, to improve computation time, we use a super-pixel representation instead of a normal pixel representation of the image. Our segmentation method furthermore utilized color, depth, and plane information to segment the object from the background and the supporting plane.
Read more.