Feature Recognition of Human Motion Behavior Based on Depth Sequence Analysis

The current research on still image recognition has been very successful, but the study of action for video classes is a challenging topic. In this work, we propose random projection-based human algorithm to address lack depth information in color (RGB frames) that not easily affected by environmental factors such as illumination and ability recognize actions along direction view. A network structure designed take obvious advantage long- short-term memory networks controlling remembering long sequences historical information. paper constituted multiple units. At same time, constructs spatial features, temporal features three stream outputs into feature matrix, whose matrix divided segments according dimension, then inputs them layer order, achieves fusion their correlation characteristics axis. Here, proposed concept batch projection operators. This basically uses much sublimitation possible improve accuracy randomly selecting several subdependencies projections defined during projection. compressed sensing design motion acceleration data low-power body area proposed, basic idea implementation process theory compression reconstruction wireless are introduced detail.