Human Activity Recognition (HAR) has gained significant importance in various fields, including healthcare, surveillance, and human-computer interaction. In this study, we comprehensively study the related work done in the past to enhance HAR system. We have also studied the various benchmark datasets that can be employed to detect diverse human activities. The basis CNNs capture distinct features and representations. Through a combination of their predictions, our ensemble model is expected to achieve superior performance on a wide range of human activities. A case study which employs pretrained CNN demonstrates the effectiveness of CNNs in improving HAR performance and highlights the potential for real-world applications. In the last section, A framework for HAR that leverages the power of ensemble Convolutional Neural Networks (CNNs) is also proposed to enhance recognition accuracy and robustness. This framework paves the way for more accurate and reliable human activity recognition systems in practical scenarios.