Multi‐dimensional data modelling of video image action recognition and motion capture in deep learning framework

摘要: In order to improve the accuracy of small-range human motion recognition in video and computational efficiency large-scale data sets, a multi-dimensional model capture image based on deep-learning framework was proposed. First, moving foreground target is extracted by Gauss mixture model, body recognised gradient histogram. At second level, dense trajectory feature deep learning are fused, according integration global encoding algorithm convolutional neural network. feature, fusion RGB tricolour taken as learning. Finally, classification network model. The simulation experiments real sets small-scale gesture show that has high for actions. addition, Imperial Computer Vision & Learning Lab behaviour set used classify experimental data. average 85.79%. can run at speed about 20 frames per second.