Application of the Radon Transform to Automatic Detection

We describe an approach to automating the detection process that is normally performed manually by a sonar operator who studies a bearing versus time ( ) beam cross-correlation (or energy) display. The approach is based on approximating an actual -plane correlogram peak trajectory with a piecewise linear fit. This is achieved by sectioning the -plane into T-second strips in time (t), taking a Radon-type transformation of the data within each strip, thresholding and clustering the transformed data to identify peaks in the "line spectrum", and then piecing together resulting straight line segments with a peak tracker. A target present decision is made by a sequential probability ratio test which is applied to likely target tracks.