3D Semantic VSLAM of Indoor Environment Based on Mask Scoring RCNN
Algorithm 3
Data association and fusion processes.
Input: feature points and semantic features on the current key frame；
Output: 3D global semantic map coordinates;
(1)
Coordinate system;
(2)
Mark unusable points on map;
(3)
Determine current frame;
(4)
if initial frames then
(5)
(1) Find map point coordinates corresponding to the target feature points;
(6)
(2) Get semantic information about the target , where is the category, is the confidence of the detection result, and is the target contour;
(7)
(3) Semantic information is associated with geometric feature points through mapping relation so that feature points have both geometric and semantic information;
(8)
(4) The relative motion of the camera is calculated according to feature matching, and the coordinates of the 3D map corresponding to the target feature points are found;
(9)
else
(10)
(5) The new parameters are substituted into the built model;
