Micro Action

Micro-action research focuses on identifying and interpreting subtle, low-intensity human movements, often conveying emotional states or intentions not readily apparent through macro-actions. Current research emphasizes developing robust methods for detecting and classifying multiple, overlapping micro-actions from video data, often employing deep learning architectures like ResNet enhanced with modules such as squeeze-and-excitation and temporal shift modules, and incorporating techniques like Dynamic Time Warping and manifold embedding. This field is significant for its potential applications in diverse areas, including improved remote physical therapy, lie detection, psychological assessment, and human-computer interaction, by providing a more nuanced understanding of human behavior.