Activity Identification Utilizing Data Mining Techniques

Abstract

We propose a novel method that, given an unknown moving object trajectory, determines which known activity type the trajectory would belong to. The proposed method utilizes various data mining techniques, including clustering, classification, and Markov model. We collect trajectories of moving objects of known activity types and build one Markov model for each activity type. Given an unknown trajectory, we compute the likelihood of this trajectory belonging to each activity type using the Markov model and the trajectory is determined to belong to the activity type that results in the highest likelihood. We use only location information of moving objects. We do not use any other information such as color, size, or shape of objects, or contextual information. We demonstrate the effectiveness of this method using trajectories of students playing two sports activities Ultimate Frisbee and volleyball. We show that the accuracy of this method is as high as 94%.