Single image pose estimation of image capture devices

摘要

Methods for image based localization using an electronic computing device are presented, the methods including: capturing a local image with an image capture device; associating metadata with the local image; causing the electronic computing device to receive the local image; causing the electronic computing device to match the local image with a database image, where the database image is three-dimensional (3D); and calculating a pose of the image capture device based on a pose of the database image and metadata associated with the local image. In some embodiments, the metadata includes at least pitch and roll data corresponding with the image capture device at a time of image capture.

主权项

1. A method for image based localization using an electronic computing device, the method comprising:
capturing a local image with an image capture device; associating metadata with the local image; causing the electronic computing device to receive the local image; causing the electronic computing device to match the local image with at least one database image, wherein the at least one database image is three-dimensional (3D); and calculating a pose of the image capture device based on a pose of the at least one database image and metadata associated with the local image, wherein
the metadata includes at least pitch and roll data corresponding with the image capture device at a time of image capture, whereinthe calculating the pose comprises:
extracting the pitch and roll data from the metadata;extracting yaw data from the metadata;matching features between the local image and the database image; andestimating a homography matrix to solve for translation of the image capture device with respect to the database image, and whereinthe matching features comprises:
extracting high resolution scale invariant feature transform (SIFT) features from the local image and the database image;performing a nearest neighbor search to establish any correspondence between the extracted SIFT features of the local image and the database image;excluding any corresponding database image SIFT features that lack depth data; andexcluding ground plane SIFT features of the database image.