We are researching and developing dynamic interpersonal distance models and real-time recognition systems for the social development training therapy of children with autism spectrum disorders (ASD). In particular, we modeled the quantitative measurements of chasing behaviors observed during therapy. Chasing behaviors are a highly social activity because children need to predict the movement of a partner (therapist). In order to measure these behaviors, the video coding using observational method by experts is needed but it is very time consuming. We consider that establishing models for real-time automated recognition of chasing behavior supports social skills development programs for children with ASD. This study focuses on chasing behavior and presents experimental results for recognition of chasing behavior during actual therapy. The proposed system reveals that it is possible to extract tracking behaviors which closely agree with therapist observations.