Beyond the Horizon: Drone-Assisted HAR Through Cutting-Edge Caps Net and Optimization Techniques

The intricate stances, the need to comprehend multiple points of view, and the varied settings in which the action unfolds are the main obstacles. In this paper, a new Capsule Network (CapsNet) is proposed as a solution to these problems. By removing the pooling layers and replacing them with capsule layers, CapsNet overcomes the major limitations of Convolutional Neural Networks (CNNs). In order to enhance the classification accuracy, the parameters of the suggested model are fine-tuned using the rider optimisation algorithm (ROA). To optimise current deep CNN designs for learning temporal information, the suggested network can be employed as a plug-in module, doing away with the necessity for a distinct temporal stream. Two freely available benchmark datasets, MOD20 and Okutama, have been used to test it. Associated to the prior state-of-the-art performances, the suggested perfect is light years ahead with an accuracy of 97.49% and 98.12% on the relevant datasets.