Activities Prediction Using Structured Data Base

Abstract

Predicting human activities is very important for human-aware robotic applications. The goal of this work is to forecast human activities that may require robot assistance. The proposed method applies the depth and visual information and the database. The activity is parsed into consecutive actions, some attributes of the actions are described by the probability functions. The method delivers the motion trajectories to nominally possible motion goals. The reasoning process is described by the graphs. The approach was evaluated using four data sets: CAD 60, CAD-120, WUT-17, and WUT-18. The solution efficiency comparing to the other state-of-art was investigated.