Performance improvement in artificial intelligence-based objects tracking via probabilistic estimation approach

Abstract

The present research is focused on the object tracking domain based upon a new probabilistic estimation approach. This is realized through a generalization of the particle filter framework (PFF), in association with a neural network, namely an intelligence-based PFF (IPFF). In this idea, a number of appropriate features of mobile objects should first be considered for use in the process of IPFF realization to make the estimation and better performance. The applicability of the proposed approach has been considered though three separated scenarios, including non-stationary, stationary/non-stationary and finally stationary objects, as long as the standard mean shift object tracking approach is realized as a benchmark approach. The experimental results verify the approach performance improvement.