Scene Understanding

Principal Investigator: Prof. Joni Kamarainen

Our research aims at total scene understanding. Relevant sub-topics are object class detection (visual object categorisation, VOC), specific object detection and event analysis from video. These are some of the most actively investigated topics in computer vision. We develop methods which can be used for efficient visual search and indexing in large scale image and video databases and Web.

Our main interest for the past few years have been part-based methods for object class representation, learning and detection. Our methods are general, but particularly good results have been achieved for face and car license plate detection and localisation. Our main scientific contribution is a novel local feature referred to as "simple Gabor feature space" or "multi-resolution Gabor feature", which provides robust and invariant feature for learning and detecting local object parts. Our main focus is now on spatial (constellation) models of local parts and on semi-supervised and completely unsupervised methods for visual object categorisation.

Featured projects

Object3D2D

The goal of this project is to provide novel solutions to one of the fundamental problems in computer vision and artificial intelligence - detection, localisation and recognition of objects in 2D and 3D (range) images. The provided methods should work similarly to the human visual system robust to real scene and imaging distortions, such as 3D pose change, illumination, occlusion and background clutter.

RTMosaic

The aim of the project is to develop real-time or near real-time methods for mosaicking and later for three-dimensional reconstruction or structure-from-motion. These methods should work on commodity computer hardware.