In this study, data collection, algorithm development, GUI and software tests were performed to develop and verify a real-time motion analysis system. First, image data were collected by photographing images of participants performing various exercise movements as a result of data collection and preprocessing, and training data was constructed by labeling joint location information. Through this, the data was systematically constructed so that the model could accurately recognize and analyze the exercise motion. Next, joint position extraction and joint angle calculation were performed using the Convolutional Neural Network (CNN) and Part Affinity Fields (PAFs) algorithms, and through this, useful results were obtained for analyzing the motion of objects and predicting the interaction. As a result of GUI development, various functions were developed to enable real-time visualization and analysis of motion data, and motion data was visually displayed as an overlay on the input image, and information such as the number of detected people, FPS, and frames was provided to provide a convenient environment to users. In the software execution results, participants' movements were detected in real-time, joint positions were accurately output, frame movements were recognized in real-time through real-time cam shooting using the OpenPose library, and the angle of the joint was also verified within the error range through an intuitive measurement app.