Object Tracking Comparison for Automated Vehicles Using MathWorks Toolsets

Object trackers are a tool to achieve accurate object state estimation over time. Due to their complexity, a framework to experiment with different variations of trackers and their subcomponents is desired. This drove the authors research and experimentation with object tracking using MathWorks toolsets. In this paper, three object trackers - Point Target Tracker (PTT), Gamma Gaussian Inverse Wishart Probability Hypothesis Density (GGIW-PHD), and Gaussian Mixture Probability Hypothesis Density (GM-PHD) - are compared in simulation for track statistics and object/track accuracy. The results show that a rectangular GM-PHD multi object tracker outperforms the other trackers. A follow up is shown using real-world data and the process used to get the sensor data into the appropriate MathWorks format. The impact of COVID-19 prevented the collection of ground truth data so the real-world data cannot be compared using the same metrics. For this reason, the simulation portion of this paper will act as the detailed discussion of fusion and tracking while the real-world testing portion is an overview of the authors’ process of converting real-world sensor data into a format compatible with MathWorks object tracking tools.