This paper describes how the data acquired from the accelerometer sensor worn by a learner in a collaborative learning situation can be utilized to guess and estimate collaborations between learners. The frequency of body movement can be used to guess the state of activity of the learner, and the difference of frequencies between two persons, i.e., synchronization of two, can be utilized to judge the similarity of their activities. We proposed using this similarity to guess the strength of ties of collaboration between two in a target period of analysis. We tested the validity of this proposed method by accelerometer data recorded in actual real-world collaborative learning projects, and the result justified our method of detecting collaboration. Also we discussed the limitation of our proposed method.