公开(公告)号：US10740607B2

公开(公告)日：2020-08-11

申请号：US15980051

申请日：2018-05-15

Applicant: AUTEL ROBOTICS CO., LTD.

Inventor： Jiangyuan Mei

IPC: G06K9/00 ,  G06K9/66 ,  G06K9/20 ,  G06K9/62 ,  B64C39/02 ,  G06N7/00 ,  G06N3/08 ,  G05D1/12 ,  G05D1/00 ,  G06N3/04

Abstract: The present invention discloses a method for determining a target through intelligent following of an unmanned aerial vehicle. The method includes: obtaining an image returned by an unmanned aerial vehicle; obtaining, according to a click of a user on the image returned by the unmanned aerial vehicle, a region picture in which the user is interested; loading a deep learning network model, inputting the obtained region picture to the deep learning network model, and outputting, by using the deep learning network model, a plurality of candidate circumscribed frames for framing targets in the region picture and probabilities that the targets in the candidate circumscribed frames belong to preset categories; and determining, according to the candidate circumscribed frames and the probabilities that the targets in the candidate circumscribed frames belong to the preset categories, whether a target image exists in the region picture. The present invention has a small calculation amount, a short operation time, and low requirements on hardware devices.