Systems and methods for edge points based monocular visual SLAM

A system comprising:

• one or more data storage devices operatively coupled to one or more hardware processors and configured to store instructions configured for execution by the one or more hardware processors to:
• acquire successive frames, in real time, from multiple views captured by a monocular camera located in a simultaneous localization and mapping (SLAM) environment and characterized by multiple unknown poses;
• detect edges in each of the successive frames;
• thin the edges of each of the successive frames to obtain thinned edges having a single pixel width;
• apply an edge filtering process on the thinned edges to eliminate noise therein;
• estimate point feature correspondences of edge points constituting the thinned edges by tracking the edge points using an optical flow;
• identify a first frame as a keyframe and further identifying a plurality of keyframes, from the acquired successive frames, based on the estimated point feature correspondences and satisfying a first set of constraints;
• perform a two-view initialization using a selected pair of keyframes from the plurality of keyframes satisfying a second set of constraints to obtain a validated 3D structure;
• estimate a trajectory of estimated poses from the multiple unknown poses of the monocular camera by obtaining incremental poses based on the validated 3D structure constituting 3D structure points;
• and
• rectify drifts in the estimated trajectory captured in the multiple views using loop closure.

A computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:

• acquire successive frames, in real time, from multiple views captured by a monocular camera located in a simultaneous localization and mapping (SLAM) environment and characterized by multiple unknown poses;
• detect edges in each of the successive frames;
• thin the edges of each of the successive frames to obtain thinned edges having a single pixel width;
• apply an edge filtering process on the thinned edges to eliminate noise therein;
• estimate point feature correspondences of edge points constituting the thinned edges by tracking the edge points using an optical flow;
• identify a first frame as a keyframe and further identifying a plurality of keyframes, from the acquired successive frames, based on the estimated point feature correspondences and satisfying a first set of constraints;
• perform a two-view initialization using a selected pair of keyframes from the plurality of keyframes satisfying a second set of constraints to obtain a validated 3D structure;
• estimate a trajectory of estimated poses from the multiple unknown poses of the monocular camera by obtaining incremental poses based on the validated 3D structure constituting 3D structure points;
• and rectifying drifts in the estimated trajectory captured in the multiple views using loop closure;
• and
• perform camera track loss recovery for continuous tracking of the monocular camera.