Estimativa de postura a partir de imagens de profundidade

Abstract

We present a method to estimate the posture of a person from depth images. We use an approach similar to that made by Grest et al. in Nonlinear body pose estimation from depth images, which use a technique to find correspondences between image and the model that will represent the posture with a nonlinear optimization technique. We use the adaptation of ICP (Iterative Closest Point) algorithm attached to the least squares method. Model construction is done automatically using Dijkstra’s algorithm to calculate distance in the point cloud. Image acquisition was performed using Kinect.