Skip to Main Content
The current method of detection of a radar target is based on the setting of a threshold determined by the average of the background returns in the region of interest. Problems arise with this method when attempting to detect small targets in littoral waters since in designing the detector it is necessary to make assumptions concerning the statistical behaviour of the background clutter. Since only long term data is available and short term prediction is required there is an inevitable missed detection/false alarm problem. The problems associated with detecting low observable targets using track-before-detect systems based on Hough transform or dynamic programming techniques are reviewed. An alternative self-adaptive spatio-temporal CFAR system and a multiple hypothesis tracker based on multiple intelligent software agents are described. The process is not perfect but, by assuming that there will be too few data measurements to establish the clutter statistics accurately, a robust sub-optimal solution is formed. The process reported is not restricted to radar returns but has potential applications in infra-red and electro-optical systems, and for processing images in particle physics and astronomy.