Trajectory Space Factorization for Deep Video-Based 3D Human Pose Estimation

Existing deep learning approaches on 3d human pose estimation for videos are based on Recurrent or Convolutional Neural Networks (RNNs or CNNs). , RNN-based frameworks can only tackle sequences with limited frames sequential models are sensitive to bad frames and tend to drift over sequences. Although existing CNN-based temporal frameworks attempt to the sensitivity and drift problems by concurrently processing all input in the sequence, the existing state-of-the-art CNN-based framework is to 3d pose estimation of a single frame from a sequential input. In paper, we propose a deep learning-based framework that utilizes matrix for sequential 3d human poses estimation. Our approach processes input frames concurrently to avoid the sensitivity and drift problems, and outputs the 3d pose estimates for every frame in the input sequence. More, the 3d poses in all frames are represented as a motion matrix into a trajectory bases matrix and a trajectory coefficient matrix. trajectory bases matrix is precomputed from matrix factorization approaches as Singular Value Decomposition (SVD) or Discrete Cosine Transform (DCT), the problem of sequential 3d pose estimation is reduced to training a deep to regress the trajectory coefficient matrix. We demonstrate the of our framework on long sequences by achieving state-of-the-art on multiple benchmark datasets. Our source code is available at: : //github. com/jiahaoLjh/trajectory-pose-3d.