Severe-Dynamic Tracking Problems Based on Lower Particles Resampling

For a target as it with large-dynamic-change which is still challenging for existing methods to perform robust tracking; the sampling-based Bayesian filtering often suffer from computational complexity associated with large number of particle demanded and weighing multiple hypotheses. Specifically, this work proposes a neural auxiliary Bayesian filtering scheme based on Monte Carlo resampling techniques, which to addresses the computational intensity that is intrinsic to all particle filter, including those have been modified to overcome the degeneracy of particles. Tracking quality for severe-dynamic experiments demonstrate that the neural via compensate the Bayesian filtering error, with high accuracy and intensive tracking performance only require lower particles compare with sequential importance resampling Bayesian filtering, meanwhile, our method also with strong robustness for low number of particles.