dc.description.abstract | This thesis gives a reliable machine model that recognizes the action of "look down at phone" and distinguishes it from other similar actions in a given consecutive video. It first reproduces facial recognition research of dimpler. Then it moves on to introduce body action recognition and explains key factors like landmarks and Openpose used in the research. It then presents the action of "look down at phone" as the research focus and briefly mentions related works to the topic. Later on, the thesis presents methods on how facial expression is performed and quickly moves on the body expression detection techniques. For body expression detection, the thesis first explains the process with image inputs, then continues to explain the process for video samples. At the end of the methods chapter, it demonstrates how the machine model processes a 26-second video with complex actions and gives a reliable and correct estimation of the actions the video contains. The thesis later presents the results of this research and compares the estimation given by the machine model to the true answers of given samples. At last, the thesis concludes the effect and benefit of this machine model and suggests future works for this research. | en |