Tracking Multiple Unmanned Aerial Vehicles on SE(3) Using a Monocular Camera

This article presents MTT- $\text{SE}\mathopen {}(3)\mathclose {}$ , a vision-based multiple-target tracking algorithm for targets that evolve on the special Euclidean group $\text{SE}\mathopen {}(3)\mathclose {}$ in dense clutter. The target state consists of the pose in $\text{SE}\mathopen {}(3)\mathclose {}$ and the twist of the vehicle. Contributions of the article include the development of a robust track initialization scheme designed on $\text{SE}\mathopen {}(3)\mathclose {}\times \mathbb {R}^{6}$ , and a track-to-track association and fusion algorithm to merge similar tracks evolving on $\text{SE}\mathopen {}(3)\mathclose {}\times \mathbb {R}^{6}$ . The performance of MTT- $\text{SE}\mathopen {}(3)\mathclose {}$ is verified in hardware by simultaneously tracking three multirotors using a stationary monocular camera. The results are compared to a multiple-target tracking algorithm that uses linear, time-invariant nearly constant velocity, acceleration, and jerk models, where it is shown that MTT- $\text{SE}\mathopen {}(3)\mathclose {}$ improves several traditional tracking metrics.