Fast Human Pose Detection Using Randomized Hierarchical Cascades of Rejectors

This paper addresses human detection and pose estimation from monocular images by formulating it as a classification problem. Our main contribution is a multi-class pose detector that uses the best components of state-of-the-art classifiers including hierarchical trees, cascades of rejectors as well as randomized forests. Given a database of images with corresponding human poses, we define a set of classes by discretizing camera viewpoint and pose space. A bottom-up approach is first followed to build a hierarchical tree by recursively clustering and merging the classes at each level. For each branch of this decision tree, we take advantage of the alignment of training images to build a list of potentially discriminative HOG (Histograms of Orientated Gradients) features. We then select the HOG blocks that show the best rejection performances. We finally grow an ensemble of cascades by randomly sampling one of these HOG-based rejectors at each branch of the tree. The resulting multi-class classifier is then used to scan images in a sliding window scheme. One of the properties of our algorithm is that the randomization can be applied on-line at no extra-cost, therefore classifying each window with a different ensemble of randomized cascades. Our approach, when compared to other pose classifiers, gives fast and efficient detection performances with both fixed and moving cameras. We present results using different publicly available training and testing data sets.