Object Tracking of Aerial Imaging Device Image Using Variational Autoencoder and External Memory

Object tracking is a fundamental problem in the field of computer vision. The object tracking methods proposed so far can be divided into a 'discriminative correlation filter' and a 'deep learning' based methods with a complex structure and a lot of computation. In this paper, we propose an algorithm for tracking objects with a simple structure while maintaining tracking performance using a convolutional variational auto-encoder, external memory, and a Siamese network. As a result of an experiment with an RT (real-time) data set to measure real-time, the result was a precision of 0.546 and a success rate of 0.527.