Deep-Learning-Based Multiple Model Tracking Method for Targets with Complex Maneuvering Motion

The effective detection of unmanned aerial vehicle (UAV) targets is great significance to guarantee national military security and social stability. In recent years, with the development communication control technology, movement UAVs has become increasingly flexible complex, presenting diverse trajectory forms different motion models in phases. Gaussian mixture probability hypothesis density filter incorporating linear jump Markov system approach (LGJMS-GMPHD) provides an efficient method for tracking multiple maneuvering targets, as applied switching motions between a set Markovian chain. However, practice, model parameters are generally unknown uncertain. When preset filtering mismatched, performance dramatically degraded. this paper, within framework LGJMS-GMPHD filter, deep-learning-based proposed. First, adaptive turn rate estimation network designed solve mismatch caused by coordinate models. Second, state modification large errors phase uncertain switching. Finally, based on simulations cluttered environments experimental field data verification, it can be concluded that proposed strong adaptability effectively improve complex motion.