PROJECT TITLE :

A Spatio-Temporal Multi-Scale Binary Descriptor

ABSTRACT:

Robotic navigation and multi-view matching are common uses of binary descriptors. However, in non-planar settings, their matching performance suffers greatly due to size and viewpoint alterations. With this difficulty in mind, we propose to use concise spatio-temporal descriptors to record the changing appearance of selected 3D scene points tracked by a moving camera. Tracking interest spots and capturing their temporal fluctuations at many scales is the first step in accomplishing this goal Feature tracks are then verified by 3D reconstruction, and the temporal sequence of descriptors is compressed using the most common and stable binary values.. Finally, using a matching method that can tolerate large scale differences, we find multi-scale correspondences between different viewpoints. If you're looking for a way to analyse a wide range of binary descriptors, this approach is ideal. Through comparisons of alternative temporal reduction strategies and the use of several binary descriptors, we demonstrate the efficiency of a combination multi-scale extraction and temporal reduction strategy