Lightweight skeleton motion classification method, system and device based on motion attention guidance and medium

• Main Authors: MIAO QIGUANG, XIN WENTIAN, LIU RUYI, WU SHUAI, LIU YI, LU ZIXIANG, ZHAO PEIPEI, LI YUNAN, WU MENGYAO, XIE KUN
• Format: Patent
• Language: Chinese; English
• Published: 14.06.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a lightweight skeleton action classification method, system and device based on action attention guidance and a medium. The method comprises the following steps: acquiring a to-be-trained skeleton video sample; constructing a lightweight skeleton motion classification model (AL-GCN) based on motion attention guidance; spatial features are extracted through an autonomous learning spatial convolutional network (AL-SGCN), and attention correction is carried out on the skeleton sequence of which the spatial features are extracted; long-distance random dependence of skeleton joints is modeled through a multi-scale time fusion convolutional network (MS-TFCN), and time feature extraction is realized; carrying out attention correction on the skeleton sequence of which the time features are extracted; using a jump module (Jump) and a multi-stream Gaussian weight selection algorithm to improve the action recognition accuracy of the AL-GCN model, and sending the generated skeleton spatial-tempora