Predicting 3D People from 2D Pictures

Variational Monte-Carlo Approach for Articulated Object Tracking

“…An "articulated" object is defined [1] as a multi-body system composed of at least two rigid components and at most six independent degrees of freedom between any components. A non-rigid, but constrained dependence exists between the components of an articulated object [2]. Examples include the human body, most animals, manipulation robots, long lorries with trailers and many others.…”

Probabilistic 3D Tracking: Rollator Users' Leg Pose from Coronal Images

“…To a lesser extent, recent literature has also focused on 3D tracking from monocular sequences (e.g. [20], [22], [28], [29]). However, these approaches often rely on full-body models for contextual cues.…”

Efficient initialization of Mixtures of Experts for human pose estimation

“…These discriminative methods differ in the organization of training set and in the runtime hypothesis selection [2], varying from Bayesian mixtures of experts (BME) [2,3,4], linear/kernel regression [5], manifold embedding [6], nearestneighbor retrieval from typical examples [7], mixture of probabilistic PCA [8], to mixture of multi-layer perceptrons [9]. We choose the BME model, because the multi-modalities in the image-to-pose distributions can be well modeled by the mixtures of experts.…”

“…We choose the BME model, because the multi-modalities in the image-to-pose distributions can be well modeled by the mixtures of experts. It has produced superior results on human pose estimation in the literature [2,3,4]. For the BME model [10,11], EM algorithm is used to estimate the parameters of both the gate network and expert network.…”

“…This happens frequently for problems with large parameter space like human pose estimation. Hence, choosing initial values is an crucial step in learning BME models, while it did not gain much attention in previous work [2,3,4].…”