A Study of the Region Covariance Descriptor: Impact of Feature Selection and Precise Localization of Target

In visual tracking, selecting the right image descriptors is critical. The popular version of descriptor is known as the covariance descriptor; however, no further studies is yet developed regarding the different methodologies for its construction. This study analyzes the contribution of diverse features of an image to the descriptor and their contribution to the detection of arbitrary targets in sequences of images, in our case: Boy, David3, Bolt and Walking2 in an image sequence. The methodology to determine the performance of the covariance matrix is defined from different sets of characteristics, and a specific combination of features is needed to develop a correlation between them. Finally, when an analysis is performed with the best set of features, F4 the target detection problem reached a performance of average, 0.77, From this experiment, it is believed that we have constructed a greater solution in choosing best features for this descriptor, allowing to move forward to the next issues such as using it on others datasets.