Brief introduction to the research problem
The project has the aim of studying and implementing algorithms of computer vision, machine learning and pattern recognition for the analysis of large datasets of images.
The final target is to accurately and automatically identify places and buildings (landmarks) in urban or industrial environment.
There are many drawbacks, that make the task of matching features between a query image and the database rather difficult:
- changes in the image resolution, illumination conditions, viewpoint;
- presence of distractors such as trees or traffic signs.
Instead, the motivations, that make the problem interesting are:
- obtain an high accuracy in the retrieval phase;
- execute a fast research;
- reduced use of memory (mobile friendly);
- work well with big data (dataset size > 1M).
- A location-aware embedding technique for accurate landmark recognition
International Conference on Distributed Smart Cameras (ICDSC), 2017 [paper] [project] [slides]
- An accurate retrieval through R-MAC+ descriptors for landmark recognition
International Conference on Distributed Smart Cameras (ICDSC), 2018 [paper] [project] [slides] [code]
- Efficient Nearest Neighbors Search for Large-Scale Landmark Recognition
International Symposium on Visual Computing (ISVC), 2018
[paper] [project] [poster] [code]
- A Dense-Depth Representation for VLAD descriptors in Content-Based Image Retrieval
International Symposium on Visual Computing (ISVC), 2018 [paper] [project]
The project is totally funded by the italian region Emila Romagna.
Federico Magliani is the supervisor for this project.