The goal of the project is to develop robust and efficient techniques for the recognition and localisation of objects in images and video sequences. This goal can be subdivided into the following aims:
- Improve the use colour information in object recognition — for example, illumination changes an shadows should be taken into account in the feature calculations.
- Combine image patch and segmentation techniques — these techniques often provide access to complementary information.
- Use models of the relative positions of interest points and segmentation regions to describe objects.
- Use optimisation methods to locate the objects in images and videos based on the configurations of interest points and segmentation regions.
- The solution to the object recognition problem is important in the following two applications. Content-based image and video retrieval: the amount of visual information that we have access to is exploding, but the techniques for searching through this visual information are not well developed. Algorithms that can recognise objects in images and videos will allow information extracted directly from an image or video to be used in locating it using a search engine.
- Video surveillance: automating the analysis of video streams from surveillance cameras will help in coping with this flood of information. Object recognition algorithms can be applied for example to recognise if an object lying in a surveyed location represents a potential threat (such as an unattended suitcase) or not (such as an empty bottle).