DenseNet121 v. VGG16: Comprehensive Performance Analysis and Accuracy Comparison in Human Activity Recognition

From surveillance to healthcare and human computer interaction, this is a crucial field in computer vision with uses. In order to enhance the quality of life, the authors of present work have aimed to investigate the efficiency of the densely linked convolutional neural network, DenseNet121, on video sequence-based human activity identification. DenseNet121’s architecture allows this model to connect all the layers directly, so enabling gradient flow and recycling features, hence increasing its capacity to learn complex temporal and spatial patterns underlying human activities. After training a model for the benchmark human action dataset, the writers got an accuracy of 89.34%. These results highlight DenseNet121’s precise identification and classification of human activities capability. Having large effects in improving automated surveillance, athletic performance optimization, and patient monitoring in healthcare situations, results suggest that DenseNet121 can be a very strong foundation towards the creation of state-of-the-modern human action detection systems for society. More research will be conducted to improve the model’s performance and generalize it into several practical settings.