Human pose estimation – OpenPose

Human pose estimation is another area of the remarkable success of the deep neural networks and has had rapid growth in recent years. In the last few chapters, we learned that deep neural networks use a combination of linear (convolution) and nonlinear (ReLU) operations to predict the output for a given set of input images. In the case of pose estimation, the deep neural network predicts the joint locations, when given a set of input images. The labeled dataset in an image consists of a bounding box determining N persons in the image and K joints per person. As the pose changes, the orientation of the joints change, so different positions are characterized by looking into the relative position of the joints. In the following sections, we'll describe the different pose estimation methods we can use.

...