Long Time Object Tracking Using Structural Features With Particle Filter

Abstract

Although long time tracking is an old research subject, it is still among the research subject actively attracting the attention of researchers and it is one of the research topic many studies conducted about. Object tracking with particle filter, known to be among stochastic methods, models dynamics related to tracking subjects by taking advantage of state space variables, implemented in this study. Presenting improvements in the measurement models used to determine the weight of the particles, a new measurement model based on structural features of similarity coefficients used by SSIM with adaptive histogram equalization and weighting center of the object has been developed in The Particle Filter. Experimental results show that the proposed measurement model in object tracking increase classical tracking performance by at least %18.59.

Keywords

• Particle Filter
• Long Time Object Tracking
• Image Similarity Metrics
• Model Free Object Tracking