Human Action Recognition Based on Transfer Learning Approach

Human action recognition techniques have gained significant attention among next-generation technologies due to their specific features and high capability inspect video sequences understand human actions. As a result, many fields benefited from techniques. Deep learning played primary role in approaches recognition. The new era of is spreading by transfer learning. Accordingly, this study's main objective propose framework with three phases for are pre-training, preprocessing, This presents set novel that three-fold as follows, (i) the pre-training phase, standard convolutional neural network trained on generic dataset adjust weights; (ii) perform process, pre-trained model then applied target dataset; (iii) phase exploits long short-term memory apply five different architectures. Three architectures stand-alone single-stream, while other two combinations between first two-stream style. Experimental results show recorded accuracies 83.24%, 90.72%, 90.85%, respectively. last achieved 93.48% 94.87%, Moreover, outperform state-of-the-art models same field.