Vision Aided Inertial Localization for Remotely Operated Vehicle (ROV)

This thesis details the implementation of a vision aided inertial localization system for a Remotely Operated Vehicle (ROV) in a controlled environment. The vision sub- system features a Raspberry-Pi equipped with the pi-camera and the inertial sub-system consists of an automotive/consumer grade MEMS-IMU operated by an Arduino board (Arduino Due). A novel PnP estimation algorithm, Linear Least Squares - Gradient SO(3) algorithm (SO(3) - PnP), is introduced for pose estimation using the vision sub-system. The position is estimated using a linear least squares approach while the ori- entation is computed iteratively using a gradient descent algorithm. The inertial sensors are used as a dead-reckoning system in between the vision measurements. The cross-talk between the vision and inertial sub-systems is established using ethernet(UDP). Sensor fusion is achieved using Kalman filters. Laboratory experiments validate the accuracy of the Vision-IMU module and enhance the possibility that they can be deployed in an un-controlled ocean environment to compute the pose of the ROV (within a target workspace) w.r.t. a given landmark.