Dual Learning Incremental Dimension Reduction for Single-View Human Motion Estimation
Wanyi Li, Jichu Ou, Jingmin Huang, Ling Zou, Yun Kuang, Xiaojie Huang, Xuan Xie · 2022
Estimating 3 dimensional (3D) human motion from monocular image sequence is a hot researching topic. This technique requires a lot of complex algorithms and models, and needs numerous samples to train. Furthermore, the 3D human motion will cost a lot time to be estimated, when some complex algorithms and models are utilized, such as convolutional neural network (CNN) and other networks. Thus, it is difficult to achieve more accurate results of the estimation. For such a problem, a new model called dual learning incremental dimension reduction model (DLIDRM) is proposed to achieve the 3D human motion estimation. With the learning of small-scale samples, it can estimate the 3D human motion more efficiently from monocular view, reduce the cost of running time, and get more accurate results.