Long Short-Term Memory Networks Based on Particle Filter for Object Tracking
Yanli Liu, Jingjing Cheng, Heng Zhang, Hang Zou, Naixue N. Xiong · IEEE Access · 2020
Due to the uncertainty of object motion, object tracking is a more difficult state estimation problem. The traditional tracking method based on particle filter has come into wide use, but it has high complexity and poor real-time performance in the process of tracking. As long as there are enough training data, the method based on deep neural network can fit any mapping well. In this paper, a structured Long Short-Term Memory Network based on Particle Filter(LSTM-PF) is proposed to learn and model video sequences with high uncertainty. This network draws on the idea of particle filter, which uses a set of weighted particles to approximate the latent variable and updates the latent state distribution through the LSTM gating structure according to Bayesian rules. We conduct a comprehensive experiment on two benchmark datasets: OTB100 and VOT2016. The experimental results show that our tracker has better performance than other trackers, which can effectively reduce the calculation redundancy and improve the tracking accuracy.