The purpose of the Pose Initialization and Propagation (PIP) system is to provide an absolute navigational solution (position, velocity, and attitude) to a moving vehicle without using GPS. This was developed as a navigation system for rocket launches in a GPS-denied environment, but it is applicable to a variety of moving vehicles. It was designed to be integrated with JPL’s Terrain Relative Navigation system as a test of the Mars Entry, Descent, and Landing (EDL) system. It was successfully used by JPL on Masten Space Systems’ Xombie vehicle in 2014.
A camera is used to determine an absolute position and orientation relative to a set of fiducial targets whose coordinates are known relative to an Earth-Centered, Earth-Fixed (ECEF) coordinate frame. This orientation is used to perform bias estimation and initialization of an inertial navigation system (INS) (i.e., a strap-down 3-axis accelerometer and gyroscope pair). Once an image is taken that contains the six targets, the PIP system must have an algorithm for extracting the pose in the ECEF frame. The first stage is the image processing that occurs in the following steps: (1) rectify image using a CAHVOR model, (2) apply canny edge detection, (3) create list of joined edges, (4) detect ellipses within joined edges, (5) match ellipses that have closely collocated centers as potential targets, and (6) reject potential targets with incorrect size, color ratio, location, etc. The second stage in the pose initialization algorithm occurs once the pixel level coordinates are determined. These coordinates, along with the surveyed ECEF location, are used with a solution to the Exterior Orientation (2D to 3D correspondences) algorithm that is known to be globally convergent. The third stage of the pose initialization algorithm is to determine the accelerometer and gyroscope biases, which is done by collecting four minutes of stationary IMU data and subtracting known components from the averaged values. The final stage is to propagate the IMU with the now estimated biases.
This provides a highly accurate INS initialization that can be performed automatically by a computer onboard the vehicle. The main advantages are the high absolute accuracy and the simplicity of hardware that must be installed on the moving vehicle [only a camera and IMU (inertial measurement unit) are needed onboard].
This algorithm is robust to low levels of contrast along with shadows and clutter in the image. The algorithm is known to fail if there are shadows or clutter directly on the targets. Note that it is important to ensure the targets are not significantly misaligned with the camera’s imaging plane so the ellipse center is as close as possible to the circle center.