An improved ET-GM-PHD filter for multiple closely-spaced extended target tracking

This paper presents an enhanced version of the ET-GM-PHD algorithm, a recently developed multiple extended target tracking (METT) technique. The original ET-GM-PHD filter tends to underestimate the target number, because the likelihood estimate in the state update process may poorly approximate the real one when targets are close to each other. The proposed algorithm addresses this drawback via introducing a new penalty strategy in estimating the measurement likelihood. Besides, Gaussian component labeling technique is adopted to obtain individual target tracks. Simulations show that for closely-spaced extended target tracking, the improved method achieves track continuity and exhibits better estimation accuracy over the original ET-GM-PHD filter.