Multi‐modal fusion method for human action recognition based on IALC

In occlusion and interaction scenarios, human action recognition (HAR) accuracy is low. To address this issue, paper proposes a novel multi-modal fusion framework for HAR. framework, module called improved attention long short-term memory (IAL) proposed, which combines the SE-ResNet50 (ISE-ResNet50) with (LSTM). IAL can extract video sequence features skeleton of behaviour. improve performance HAR at high semantic level, obtained are fed into couple hidden Markov model (CHMM), IAL+CHMM method IALC developed based on probability graph model. test proposed method, experiments conducted HMDB51, UCF101, Kinetics 400k, ActivityNet datasets, 86.40%, 97.78%, 81.12%, 69.36% four respectively. The experimental results show that when environment complex, achieve more accurate target results.