Tracking Untrained Objects Based On Optical Flow Approach

초록

As Computer Vision is being used in various fields these days, the fields that require object tracking increase, and the performance starts to improve accordingly. In object tracking, State of the Art models are methodologies that obtain high accuracy by learning the objects to be tracked, whereas they have low accuracy for unlearned objects. To compensate for this, some models provide an environment for transfer learning on the object to be tracked, but there is rarely a large amount of preprocessed data that can be trained. In such a problem situation, it is possible to track an object universally and accurately using Lukas Kanade optical flow and additionally derive center coordinates.

키워드