Global Energy Minimization and Optimization of Multi-Target Tracking

Abstract

The recent multiple target tracking methods aim to obtain the best possible number of trajectories within the time frame, and few constraints have been set to handle the wide area of trajectories by discrete mapping. In this novel approach of multi target tracking, energy terms are formulated to attain the global optimization which includes the entire representation of the issues such as target tracking, operational representation, collision handling and trajectory processing. Furthermore, two optimization strategies such as the gradient descent which is performed on multiple feature space to obtain local minima of a density function from the given sample of data and gradient ascent which is carried out to achieve a likelihood matching of the target and used to handle the partial evidence of the image, and also uncertainty of the various targets are minimized. The experimentation is performed on the openly available dataset and the mean target tracking accuracy and precision is studied to validate the proposed tracking.