Spatio-temporal Whitening of Imaging Sensor Data Streams Using Three-Dimensional Linear Prediction

A three-dimensional whitening filter based on linear prediction is described. It is designed to remove textured backgrounds, with spatial and temporal correlation, in data streams acquired from imaging electro-optic sensors operating in visible, Ultra-Violet (UV) or Infra-Red (IR) bands. The background is modeled as an Auto-Regressive (AR) process. The operation of the filter is examined using real images of ocean waves obtained using a video camera with a synthetic, dim, sub-pixel target, inserted. Optimal filter parameters are determined using Receiver Operating Characteristics (ROCs). The output of the whitener is filtered using a Probabilistic Data Association (PDA) filter with automatic track initiation. The whitener/tracker combination yields an average true-track confirmation delay of 0.303 s when the average false-track confirmation rate is 1.33 s^-1, for a barely visible point-target that is 10% brighter than the background pixel that it replaces.