Over the years, deep learning models have been applied to human action recognition (HAR). due to the enormous amount of labeled data needed to train deep learning models, there has been a significant delay in the absolute development of these models. Data collection in sectors like HAR is challenging, and human labeling is expensive and time-consuming. The current approaches mainly rely on manual data gathering and accurate data labeling, which is carried out by human administration. This frequently leads to a slow and prone to human bias labeling data collection method. To solve these issues, we offered a novel approach to the current data collection techniques [1]. It is generally used that (CNN) is among machine learning models. Since Yann Lecun created this context in 1988, image identification has greatly improved. Transfer learning in image classification has simplified the process of training new models from the beginning and has reduced the number of data points that need to be processed, it was used in this project to classify human actions.