Towards Optimum Drones Track-to-Track Adaptive Fusion Using LSTM

This paper proposes a track-to-track fusion of drone tracks based on the LSTM (Long Term Short Term Memory) network. The sensors are unbiased radars with fixed locations. The LSTM in a multi-stage form is used to learn the generation of fused tracks based on an input of two tracks. Since it is an online and temporal process we call it dynamic, or in artificial intelligence terms, adaptive, unlike the standard baseline approach that uses fixed weighting parameters. The technique uses an objective function utilizing performance metrics such as SIAP and OSPA. Simulations are done for multiple objects in the presence of two radars to demonstrate the validity of the proposed technique. The JPDA (Joint Probability Data Association) with fixed gating and moderate clutter is also used. The performance of the proposed method is compared with a fixed fusion weighting parameter. A series of seven scenarios with 5 runs for each scenario is implemented to have a comprehensive testing suite. Stone Soup was chosen as the radar simulation environment. Based on the standard performance metrics, it was evident that the LSTM fused approach has an upper edge over the standard equal weights fusion.