Video-based behavior recognition has emerged as a pivotal tool in clinical research for disability detection and prediction, as well as in neuroscience studies. However, current methods for behavior recognition in non-human primates (NHPs) are often labor-intensive and lack standardized evaluation protocols. In this study, we utilized tracking system and deep learning algorism to analyze drug-induced behavioral responses in Marmosets. Utilizing the [Panlab] SMART 3.0 Video Tracking Software, we successfully monitored and quantified the behavioral changes elicited by drugs such as Cocaine, Tetrahydrocannabinol (THC), and 25H-Nbome. Our approach using the snapshot method allowed for accurate tracking by separating background from marmoset activity. Additionally, utilizing Labgym, an open-source computational tool, we are developing a model to further characterize drug-induced behaviors, particularly focusing on stereotyped behavior like head shaking induced by Cocaine in Marmosets. This innovative system not only offers insights into primate behavior but also sheds light on the effects of drugs, benefiting both clinical and neuroscientific research communities with its comprehensive and insightful findings.