3-D environment mapping systems and methods of dynamically mapping a 3-D environment

摘要

A 3-D environment mapping systems and methods of dynamically mapping a 3-D environment includes identifying at least five point correspondences associated with image frames of a scene recorded by a camera from a plurality of viewpoints. A down direction is determined from IMU motion data. The down direction is an absolute global downward direction. Components of the scene that are vertical are identified using the down direction, an iteration of 3-D position of each of the five point correspondences, and a current iteration of refined relative locations and poses of the camera. A 2-D map is generated of the scene by mapping the identified components of the scene in an eagle view of the scene.

主权项

1. A 3-D environment mapping system to separate areas in an indoor scenario, said system does not include a global positioning system (GPS) tracker and does not use a priori information, said system, comprising:
a camera to generate a video stream comprised of a plurality of image frames including a scene taken from a plurality of viewpoints, each of said plurality of image frames sharing at least five point correspondences with at least one of the other of said plurality of image frames; an inertial measurement unit (IMU) associated with said camera to provide at least six axes of image motion data during recording of a plurality of said plurality of image frames; and a non-transitory computer readable medium having stored thereon a plurality of processor executable instructions that, when executed by said processor, cause said processor to perform at least one iteration of the following, with each of said at least one iteration employing at least one of a combination of point correspondences that has not been analyzed in any preceding of said at least one iteration and a combination of said plurality of viewpoints that has not been used in any preceding of said at least one iteration, identify at least five point correspondences associated with a current iteration present within at least two of said plurality of said plurality of image frames, determine a current iteration unfused relative locations and poses of said camera by analyzing said at least five point correspondences associated with said current iteration, determine a current iteration three-dimensional (3-D) position of each of said at least five point correspondences associated with said current iteration, produce a current iteration's version of a refined relative locations and poses of said camera by applying a data fusion algorithm to said current iteration unfused relative poses and locations and said motion data, determine a down direction from said IMU motion data, wherein said down direction is an absolute global downward direction with respect to the Earth, identify components of said scene that are vertical using said down direction, said current iteration, and said current iteration's version of said refined relative locations and poses of said camera, and generate a current iteration's version of a two-dimensional (2-D) map of said scene for an indoor scenario, by mapping said identified components of said scene that are vertical into an eagle eye view of said scene, said eagle eye view of said scene showing said identified components that are vertical for said indoor scenario, wherein said identified components that are vertical are vertical surfaces for said indoor scenario, said eagle eye view of said scene showing separate areas of said indoor scenario defined by said vertical surfaces.