LM-Net: a dynamic gesture recognition network with long-term aggregation and motion excitation

• Published in: International journal of machine learning and cybernetics 2024-04, Vol.15 (4), p.1633-1645
• Main Authors: Chang, Shaopeng, Huang, Xueyu
• Format: Article
• Language: English
• Subjects: Artificial Intelligence; Complex Systems; Computational Intelligence; Control; Deep learning; Engineering; Excitation; Gesture recognition; Harnesses; Machine learning; Mechatronics; Modules; Movement; Neural networks; Original Article; Pattern Recognition; Robotics; Systems Biology
• ISSN: 1868-8071; 1868-808X
• DOI: 10.1007/s13042-023-01987-3

In recent years, there has been a growing interest in dynamic hand gestures as a natural means of human–computer interaction. However, existing methods for recognizing dynamic gestures have certain limitations, particularly in consistently capturing and focusing on the hand movement region across various motion patterns. This research paper presents LMNet, an innovative and efficacious network comprising the Long-term Aggregation Module and the Motion Excitation Module. The Motion Excitation Module exploits motion information extracted from neighboring frames to amplify motion-sensitive channels, while the Long-term Aggregation Module harnesses dynamic convolution to assimilate temporal information from diverse motion patterns. Rigorous experimentation conducted on the EgoGesture and Jester datasets demonstrates that LMNet surpasses the majority of prevailing approaches in terms of accuracy, while concurrently upholding an optimal computational cost.