Target Tracking Using a Mean-Shift Occlusion Aware Particle Filter

Most of the sequential importance resampling tracking algorithms use arbitrarily high number particles to achieve better performance, with consequently huge computational costs. This article aims address problem occlusion which arises in visual tracking, using fewer particles. To this extent, mean-shift algorithm is incorporated probabilistic filtering framework allows smaller particle set maintain multiple modes state probability density function. Occlusion detected based on correlation coefficient between reference target and candidate at filtered location. If detected, transition model for switched a random walk enables gradual outward spread larger area. enhances recapturing post-occlusion, even when it has changed its normal course motion while being occluded. The likelihood built combination both color distribution edge orientation histogram features, represent appearance structure, respectively. evaluated three benchmark computer vision datasets: OTB 100, xmlns:xlink="http://www.w3.org/1999/xlink">VOT 18 xmlns:xlink="http://www.w3.org/1999/xlink">TrackingNet . performance compared fourteen state-of-the-art algorithms. From quantitative qualitative results, observed that proposed scheme works real-time also performs significantly than state-of-the-arts sequences involving challenges fast motions.