A GMM-based multitarget tracking algorithm and analysis

In this paper, we propose a Gaussian mixture (GM)-probability hypothesis density (PHD) filter based algorithm for multiple objects tracking. To reduce the number of used Gaussians, we introduced a clustering procedure and observation Gaussians estimation, to avoid the exponential growth of mixture components when the number of measurement highly increases. The new birth of Gaussian components is performed adaptively by selecting the measurements that did not update any significant component from previous time step. Simulation results have proved the effectiveness of initiation, survival and termination of tracks using our proposed technique. The components number of the updated mixture and the execution time are reduced significantly.