ego-motion Precise ego-motion estimating deep network through rearrangement of image sequences which can also allow the prediction of an automobile driving in reverse

摘要

The present invention relates to a method for increasing accuracy of ego-motion of a deep network and learning to drive in reverse of an automobile using rearrangement of image sequences. The method comprises the steps of: obtaining frames without limiting frame intervals of a camera sensor to frame to frame; rearranging the obtained frames to increase data; and inversely learning the obtained frames and generating small loop closing using a plurality of sequences.