Object Tracking using Statistic-based Feature Fusion Technique

Object tracking in wide area motion imagery (WAMI) is challenging because of many factors including small target sizes, viewpoint changes, object rotation, occlusions, and shadows. One method to overcome these challenges is to fuse multiple features of different types to obtain a robust understanding of the object. For multi-feature fusion based tracking applications, the weighting of the features will highly affect the outcome. While obtaining a constant weighting scheme based on training sequences is possible, an adaptive method may better utilize the features. An adaptive weighting scheme should favor the most discerning features in the previous frames. A known way to determine a feature’s ability to discern the target from the background is based on statistics analysis. We propose to use the statistics-based fusion method to better utilize rotation invariant based features to track objects. The effectiveness of the fusion method will be compared to a constant weighting scheme on eight sequences in two WAMI datasets.https://ecommons.udayton.edu/stander_posters/1808/thumbnail.jp