Human Activity Recognition is turning into a well-known field of exploration over the most recent twenty years. Understanding human conduct in pictures gives helpful data for a huge number of PC vision issues and has numerous applications like scene acknowledgment and posture assessment. There are different strategies present for action acknowledgment. In this postulation, methodology for human movement acknowledgment utilizing an individual's posture skeleton in pictures is proposed. This work is separated into two sections; a single individual pose assessment and activity classification. Pose Estimation comprises the acknowledgment of 18 body key points. Then, action grouping task is performed by utilizing various calculated relapses. Additionally, correlation between different other regression and algorithms' accuracy on dataset is shown. To assess the error during the pose assessment we utilize the Euclidean distance equation which brings about a normal inexact mistake of 7.166. For the action characterization task, we utilized six different supervised machine learning algorithms in which Decision Tree and Random Forest give the most noteworthy exactness of 96% and Naive Bayes gives minimal precision of 42.66%. For dataset prerequisites, we have made our own dataset with the assistance of the Roboflow system. Arrangement of dataset is done by separating into two sections, one is utilized to prepare the model and one more is utilized to approve our proposed model's presentation. We will likewise examine a few benefits and hindrances came across project and view a concise examination of exactness. The discoveries will likewise show how the vision-based methodology is turning into a well-known methodology for HAR research nowadays.