Structured Deep Learning Supported with Point Cloud for 3D Human Pose Estimation
Abstract
In this paper, a structural-output is obtained to estimate 3D human pose using 3D human point cloud and monocular images. The Neural Network takes a human image and 3D pose as inputs and gives outputs a score value. Conditional Random Field (CRF) approach is using to semantically classify human limbs in its point cloud for 3D human pose production. The voxel cloud connectivity segmentation (VCCS) is used as the segmentation method that voxelisation of the 3D point cloud. The network structure consists of a convolutional neural network for image feature extraction and pose into a joint embedding. The score function is calculation from the dot-product between the images and pose embeddings which is high when the image-pose pair matches and low otherwise. Image-pose embedding and score function are jointly trained using the max-margin cost function. Finally we present visualizations of the image-position placement field, showing that the network has learned a high level embedding of body orientation and pose configuration.
Keywords
Cite this article
Özbay, E., Çınar, A., & Güler, Z. (2017). Structured Deep Learning Supported with Point Cloud for 3D Human Pose Estimation. In International Symposium on Multidisciplinary Studies and Innovative Technologies (pp. 304–309). SET Technology.
Article History
- Published
- 8 Dec 2017
License
© 2017 SET Technology. This article is distributed under the terms of the CC BY 4.0.