Multi-object Tracking using Particle Swarm Optimization on Target Interactions

In this work, a particle swarm optimization based algorithm for multitarget tracking is presented. At the beginning of each frame the objects are tracked individually using highly discriminative appearance models among different targets. The task of object tracking is considered as a numerical optimization problem, where a particle swarm optimization is used to track the local mode of the similarity measure. The objective function is built on region covariance matrix and multi-patch based object representation. The target locations and velocities that are determined in such a way are further employed in a particle swarm optimization based algorithm, which refines the trajectories extracted in the first phase. Afterwards, a conjugate method is used in the final optimization. Thus, the particle swarm algorithm is utilized to seek good local minima and the conjugate gradient is used to find the local minimum accurately. At this stage we optimize complex energy functions, which represent the presence, movement and interaction of all targets in sequence of recent frames within a temporal window. The algorithm has been evaluated on publicly available datasets. The experimental results show performance improvement over relevant algorithms.