Autonomous navigation for underwater robots is often approached from the integration of sophisticated imagery and dead-reckoning sensors to achieve a practical implementation. Image processing has become popular in robotics applications for autonomous navigation from object recognition, visual odometry to visual Simultaneous Localization and Mapping (vSLAM). In this paper, a minimal instrumentation setup is proposed towards autonomous navigation for underwater robots based on monocular ORBSLAM2 and Point Cloud Processing in structured environments. ORBSLAM2 is a vSLAM algorithm that generates a point cloud map of Oriented FAST and Rotated BRIEF (ORB) features from video images and localizes the video source in it. We evaluate the feasibility of the point cloud processing in the implementation of a basic navigation method. The cloud is processed to abstract the surroundings into 3D planes where a trajectory (i.e. transects) is easily generated, and a way-point robot controller is capable of driving the robot given the current location as feedback. The basis of the method is tested in a simulated environment.