Compact visual features for video matching
Interest point based methods for object recognition are commonly used and proven to be robust. Recently, different methods for coding these features have been proposed, in order to reduce the size of descriptors and to match them at lower computational cost. These methods are mostly applied to still images or key frames of videos, not exploiting temporal redundancy in video. This work shall perform experiments to assess how temporal redundancy can be best used to get compact descriptors for video, compatible with existing frameworks.
C/C++, basic image processing knowledge