Paper Title
Object Localization using ORB SLAM 2 and Semantic Segmentation
Abstract
In this paper, we propose a system for localizing a target object using the integration of ORB SLAM2 and
semantic segmentation algorithms. Our system runs on Robot Operating System (ROS) message bus and utilizes an RGB-D
version of ORB slam2 to accurately estimate camera poses. Object classification is precisely predicted for each pixel in the
frame using a PSPNet model trained on ADE20K dataset and fine-tuned on SUNRGBD dataset. The resulting semantic map
assigns pre- defined semantic colors to objects in the environment. We then developed an Octomap localization algorithm
that uses these pre- defined colors to locate the target object in the environment.
Keywords - OctoMap, RGBD, Object classification