We are interested in semantic spatial cognition. This is a complex process and involves many processes.
Place recognition and learning: The goal here is to understand how places are represented, detected and recognized. Appearance plays key role in place detection and recognition as geometric or odometric data may not be always available. For this, we have developed the Topological Spatial Cognition (TSC) model (Karaoguz & Bozma, 2016).
Illumination Invariance: Changing appearances have proven to be a significant challenge in this area. Our next goal is to increase the robustness of the model against appearance variability as much as possible. One of the primary causes is due to changing illumination and lighting conditions. Therefore, we are working on neutralizing the images so that we can benefit from most of the naturally uninformative image by processing. Beside uninformative images, our second goal is to recognize places regardless of illumination conditions and hence a robot can recognize same place day and night. BD(Bubble Descriptors) and SSG (Segment Summary Graphs) are used for place detection and our performance evaluation are based on these algorithms.
For more info, pls contact Berkan Höke.
Semantic place representation: Can a robot understand its surroundings semantically? This is the goal of this project. This is again a complex task – as it requires the robot to look around in the scene and label the objects.
For more info, pls contact Doğan Patar.
This project is supported in part by Tübitak-EEEAG.