Human Activity Recognition (HAR) is a fascinating process that involves identifying and categorizing human activities based on observations of subject behavior and environmental factors. Out of the major phases of HAR, preprocessing and feature extraction, in particular, require a lot of attention because they constitute the foundation for the training phase. The current evaluation effectively examines the trends in vision-based research approaches. Many models based on the artificial intelligence are suggested for the activity recognition; however, they couldn't perform satisfactorily on real world long-term HAR due to their lacuna in the extraction of spatial and temporal data. In light of these drawbacks, we propose a hybrid methodology for the human activity recognition which integrates Convolutional Neural Network (CNN) with the Long Short-Term Memory (LSTM), where CNN works for extracting the spatial characteristics and LSTM is used for the learning of the temporal information. A complete evaluation of the substantial work done in HAR domain and its vision-based techniques would help the readers.