Gradient-Descent-Based STDM-VSIMM Filter for Maneuvering Target Tracking

In this article, we propose a maneuvering target tracking algorithm based on the sojourn-time-dependent semi-Markov variable structure interacting multiple model (STDM-VSIMM) filter. Due to the frequent changes of target motion patterns, traditional IMM algorithms with constant mode transition probability and fixed structure model sets are difficult to adapt effectively. The motion pattern sojourn time in the semi-Markov chain-driven IMM algorithm obeys a random distribution function with memory. Consequently, the transition probability between patterns depends on the elapsed sojourn time. This dependency enables the model to more accurately reflect the target’s actual motion behavior. In addition, Kullback–Leibler divergence is used as a model matching measure to dynamically adjust the model set structure and further enhance the model adaptability. Considering that the performance of tracking algorithms and data fusion often depends on parameters specific to the sensor system, we introduce a gradient descent algorithm to automatically optimize key components, including model selection threshold, process and measurement noise covariance matrices. Finally, simulation results confirm the superiority of this approach in complex aircraft maneuvering target tracking scenarios compared to traditional methods.