Skip to Content

Hierarchical Multi-Parametric and Multi-Resolution Strategy (HMPMR) for Tracking

A strategy for improving the object tracking problem using direct methods is proposed. It is focused on being robust under partial occlusions, large frame to frame motions, vibrations, and 3D changes, accomplishing the tracking task at real-time frame rates.

The algorithm is based on a hierarchical strategy in terms of image resolution and number of parameters estimated in each resolution, that we call H_MPMR.

Different numbers of parameters are integrated in the estimation of the motion model: translation (2), rigid transform (3), similarity (4), affine (6), and homography (8), and the use of these models inside a multi-resolution framework is analyzed.






 The proposed strategy is tested with data from real  light tests using a UAV, where the requirements of direct methods are easily unsatis ed due to vehicle vibrations and the problems caused by outdoors operations. The performance of the proposed strategy is analyzed by comparing it with a Multi-Resolution approach (MR) using the ICIA technique, and tracking di fferent templates from di erent camera con gurations (downwards and forwards looking con gurations).

Results show that a H MPMR approach is able to tolerate long and strong frame to frame changes (translation, scale, perspective). The results also show that the proposed strategy allows to achieve a robust tracking not only in terms of the image motion that can be handled, and the tolerance to partial occlusions, but also in terms of the speed reached when estimating the motion models that strengthen the use of the proposed strategy in other elds: mosaicing, pose estimation, registration, motion detection, etc.