Two-dimensional (2D) multi-person pose estimation and three-dimensional (3D) root-relative pose estimation from a monocular RGB camera have made significant progress recently. Yet, real-world applications require depth estimations and the ability to determine the distances between people in a scene. Therefore, it is necessary to recover the 3D absolute poses of several people. However, this is still a challenge when using cameras from single points of view. Furthermore, the previously proposed systems typically required a significant amount of resources and memory. To overcome these restrictions, we herein propose a real-time framework for multi-person 3D absolute pose estimation from a monocular camera, which integrates a human detector, a 2D pose estimator, a 3D root-relative pose reconstructor, and a root depth estimator in a top-down manner. The proposed system, called Root-GAST-Net, is based on modified versions of GAST-Net and RootNet networks. The efficiency of the proposed Root-GAST-Net system is demonstrated through quantitative and qualitative evaluations on two benchmark datasets, Human3.6M and MuPoTS-3D. On all evaluated metrics, our experimental results on the MuPoTS-3D dataset outperform the current state-of-the-art by a significant margin, and can run in real-time at 15 fps on the Nvidia GeForce GTX 1080.

In this paper, we propose a new spherical mixture model for head detection in depth images. The shape of objects is considered through the statistical distribution of points in three-dimensional space. We first introduce a new probability density function belonging to the family of elliptical distributions and designed to model points spread near a spherical surface. The unknown parameters of the mixture model based on this new density are estimated with the expectation-maximization algorithm. The properties of the proposed estimators are studied from both the theoretical and practical points of view. We show the effectiveness of the proposed method for the problem of head detection. Results are compared to those obtained with RANSAC based methods.

People detection is a crucial component in many applications such as stadium or building evacuation and marketing analysis. It gives important information on the current number of people in a building or a room. In this work, we propose an algorithm for a real-time people detection system from a sequence of depth images acquired with a vertical depth sensor. The processing pipeline includes a pre-processing and a people segmentation module. The proposed algorithm handles specific situations such as missing or truncated heads. The efficency of the method is evaluated on a set of depth image sequences.

Image inpainting consists in reconstructing missing parts of a given image. In this work, we propose to use an approximation method based on splines and finite elements approximation to recover missing depth values in images acquired with a Kinect 3D sensor. Neighboring pixels in the depth map may contain very different distance values. The considered surface approximation problem therefore involves rapidly varying data which can lead to oscillations (Gibbs phenomenon). To address this issue, we propose to apply two scale transformations to dampen these oscillations near steep gradients implied by the data. The algorithm is presented with some numerical examples of inpainting. Our approach also permits to get a finer resolution of the 3D depth map.