23 Real Time Human Detection using Covariance Matrices as Human Descriptor

Tracking the moving objects in video and classifying them as human or nonhuman object is an important problem in computer vision. We present real time human detection system utilizing covariance matrices as object descriptors. We describe a fast method for computation of covariance based on integral images. The idea presented here is more general than the image sums or histograms. Covariance matrices do not lie on Euclidean space, therefore we use Logitboost classifier modified for analytic manifold for classification. The algorithm is tested on CAVIAR human database where superior detection rates are observed over the previous approaches.