作者: Niall McLaughlin , Jesus Martinez del Rincon
DOI: 10.1007/978-3-319-94544-6_2
关键词:
摘要: In this paper, a discriminative human pose estimation system based on deep learning is proposed for monocular video-sequences. Our approach combines simple but efficient Convolutional Neural Network that directly regresses the 3D with recurrent denoising autoencoder provides refinement using temporal information contained in sequence of previous frames. architecture also able to provide an integrated training between both parts order better model space activities, where noisy realistic poses produced by partially trained CNN are used enhance autoencoder. The has been evaluated two standard datasets, HumanEva-I and Human3.6M, comprising more than 15 different activities. We show our can state art results.