Enhanced Multiple Target Tracking Using a Generalized Multi-Target Smoothing Algorithm With Tracklet Association

Multiple Target Tracking (MTT) is an extensively researched field with significant importance and a wide range of applications. However, in challenging scenarios where the targets are closely spaced, and the complexity is high, many existing MTT methods often struggle to accurately distinguish and track individual targets. To address this issue, we propose an effective smoothing strategy. Our approach extends the Rauch-Tung-Striebel technique to handle multiple targets while also employing tracklet association techniques to manage dense multi-target scenarios. This strategy involves designing a smoothing multi-target distribution model using a Bayesian approach that utilizes both kinematic and identification information about multiple targets. It can be applied to all widely used MTT algorithms in the forward filtering step. The recursive smoothing algorithm we developed for the backward filtering step enhances inter-subject discrimination and improves track quality. Consequently, we achieve enhanced trajectories by integrating orbital correlation and smoothing techniques, especially when each trajectory is entangled with nearby objects. In this paper, we demonstrate a backward smoothing strategy tailored for a linear Gaussian model and present experimental results using infrared imaging that show improved tracking performance. Additionally, we illustrate its superiority over existing smoothing algorithms.