details
Object recognition, one of the major interests in computer vision, has a wide range of potential applications, from robotics, assisted living to surveillance, and more recently augmented reality. However, object recognition is arguably the most challenging problem in the computer vision and machine learning community due to several real world issues, e.g., occlusions, scene clutter, affine transformations, light changes, never-before-seen objects, non-rigid objects, non-textured objects, non-lambertian objects and scalability to the number of known objects. Moreover 3D object recognition requires accurate 3D models reconstructed from real objects.
Our computer science team is committed address these open problems in the state-of-the-art and develop object recognition solutions for everyday tasks. This project includes: developing real-time object recognition methods [1], designing novel image features [3], evaluating 2D and 3D object descriptors [1,3], optimizing classifiers, combining different feature-descriptor-types [2] , collecting object datasets, reconstructing 3D models and scene processing.
REFERENCES
[1] Proença, P.F., Gaspar F., and Dias, M. (2013). Good appearance and shape descriptors for object category recognition. In International Symposium on Visual Computing.
[2] Proença, P.F. (2013). Object Category Recognition through RGB-D Data. In MSc Thesis, ISCTE-IUL.
[3] Bastos, R., Dias, M.S. (2009). FIRST – Fast Invariant to Rotation and Scale Transform. VDM Verlag Dr. Müller e.K. ISBN: 978-3-639-17489-2