Many cart- and vehicular-based unexploded ordnance (UXO) detection systems employ global positioning system (GPS) receivers to accurately determine the system's position. However, the unevenness of the terrain often causes the system to tilt during the data collection, introducing errors in the GPS measurements and diminishing discrimination performance.
The objective of this SERDP Exploratory Development (SEED) project was to correct errors in GPS measurements recorded by a cart-mounted UXO detection system. The UXO detection system includes electromagnetic induction (EMI) and magnetometer sensors in conjuction with a GPS receiver, which is intended to accurately determine the position of the system mounted above the cart.
This project developed algorithms to correct for errors in the GPS measurements introduced by the tilting of the cart-mounted UXO detection system. Three approaches were considered: low-pass filtering (LPF), linear predictive filtering, and adaptive filtering using a hidden Markov model (HMM). The LPF is the baseline error correction algorithm and removes dramatic and unrealistic jumps in the GPS measurements even though it does not explicitly model the system motion. The linear predictive filter offers a causal approach, but again does not explicitly model the system motion. The third approach—the HMM approach—does explicitly include the system movement such that the cart motion is broadly characterized as either linear or nonlinear, after which an appropriate filter is applied.
All three error correction techniques were applied to simulated data for which both the sources of error and the ground truth were known so that the performance of the algorithms could be compared. The algorithms were then applied to measured data collected with a cart-based system at the Advanced Technology Demonstration at Jefferson Proving Ground (ATD/JPG) to evaluate their robustness under real conditions. Finally, in order to address deficiencies in the HMM approach, a quaternion formulation was proposed as a possible means of tracking and processing the cart orientation as it moves over the uneven terrain.
Decreasing the error in the GPS measurements tends to decrease the error in the parameter inversion, with the degree of improvement dependent upon the target parameters and orientation relative to the data collection. This observation motivates the need for techniques to reduce GPS measurement errors. The results of this project may lead to more sophisticated processing algorithms for the discrimination of UXO from surrounding metallic clutter. (SEED Project Completed - 2003)