Submarine Detection (Network Level)

In a sparse undersea sensor network, the sensor coverage areas are often non-overlapping. This type of surveillance network can achieve a reasonable target detection accuracy while minimizing the overall system cost. Although targets in the surveillance region may not always be detected, as they move, sensors can collectively detect, classify and track them. The traditional instantaneous target detection methods are less effective since, at certain sampling instances, targets may not be detected by any sensors at all! Therefore, operator must evaluate the sensor reports over a period of time. Since it is inevitable for sensors to generate false reports, over the entire evaluation period, both positive and false reports co-exist, making it difficult, if not impossible, for the operator visually or computer autonomously to identify the tracks. To overcome these problems, we have developed new temporal and spatial fusion methods for multiple targets detection and tracking. An optimization based fusion method has been developed to fuse spatially distributed sensor reports without making any assumptions of their underlying statistical distributions. The mathematical morphological operations are used to eliminate the isolated reports, thus, reducing the impact of random false detections. Our temporal fusion method further estimates the target tracks based on kinematic trajectories of moving targets. The synthetic target tracks are also estimated to provide the operator with a better view of the surveillance region.