Social behavior is a core subject in public space research, yet its quantitative representation remains a significant challenge. With the advancement of machine vision technology, automated tracking and quantitative collection of spatial behaviors have become achievable. However, computational analysis and quantitative description methods for micro-behavioral data are still lacking. This study introduces a machine vision-based behavior computation method using pose estimation algorithms, aiming to develop a workflow for the quantitative analysis of micro-behaviors in public space scenarios. The feasibility of video-based micro-behavior computation tools is demonstrated. Focusing on micro-gait analysis as an example, this research reconstructs behavioral sequences and conducts empirical experiments to determine the optimal average shooting angles and distances, thereby improving the accuracy of data collection and behavior analysis. Additionally, a Grasshopper module is developed for behavioral sequence computation, enabling the visualization of gait metrics for micro-behaviors based on 17 joint data points. The accuracy of the machine vision-based behavior reconstruction method is further validated through motion capture experiments. Finally, using a social distancing experiment as a case study, this study provides preliminary evidence that the proposed method enhances the precision of micro-behavior analysis in complex public spaces. It demonstrates the potential to effectively represent public space indicators, such as social distancing.