ABSTRACT
A parallelized region proposal network-based ORB feature extraction technique is proposed in which a significantly smaller number of bounding boxes are produced on the currently rectified stereo image and the next stereo frame for selective extraction of features and then matching. The optical flow field highlights how feature descriptors tend to diverge from the focus of expansion. This chapter uses the term PRN-SORB, which is an abbreviation for parallelized region proposal network-based Swift ORB. The development of the proposed PRN-SORB-SLAM-i.e., parallelized region proposal network-based swift ORB SLAM-which utilizes a stereo-vision setup to project a trajectory for local path planning is elaborated in detail in the chapter. The depth perception and formation of a fused image from a given stereo image pair work on the triangulation principle. A solution, however, is to make use of bundle adjustment.
