Object Detection and 6D Pose Estimation for Precise Robotic Manipulation in Unstructured Environments

In this paper we present an algorithm for the robust 6D pose estimation with an RGB-D camera in harsh and unstructured environments using object detection. While the pose estimation uses clustering and segmentation to find a robust point in multiple frames to track changes in the position of the camera, its functionality is enhanced with Faster-RCNN for classification and detection, providing the operator with information about the object of interest. This work further facilitates the goal of increasing the robot's autonomy and helping operators to recover 3D reconstructions of the objects to be manipulated with the robot.