Interactions between people are of utmost magnitude for cross-device systems development. By using this kind of software, devices owned by those people end up interacting between themselves, and, therefore, making the system work. This work proposes to elaborate a toolkit that can detect and analyze those human interactions by using computer vision over videos showing them. All of these through the usage of 3D modeled test scenarios in addition to applying proxemics metrics and concepts of F -formations patterns so we can define them at various interaction types. To meet this goal, we used a previously trained human detection model in conjunction with two proposed concepts to estimate indispensable values: Distance between people, their body orientation, and relative position. To validate this tool, we tested it with a hundred test cases, each one having a set of different F -formation types so we could get the effectiveness of its detection functionality.