The application of Smart Environment plays an important role in the development of advanced science and technology and therefore more and more attention. And activity recognition is the basis of Smart Environment, which reflects the intelligence of Smart Environment. However, there are two difficult and important problems which limiting the popularization of Smart Environment existing: high costs and difficulties in obtaining activity pattern. In order to overcome these problems and obtain activity pattern more effectively and efficiently, a framework for activity pattern transfer is proposed in this paper. There are two parts of activity pattern transfer: (i) Trajectory transfer, establishing the relationship on trajectories of template environment and new environment. (ii) Trigger duration transfer, transferring the trigger duration from template environment to new environment. There are four core algorithms of activity recognition based on transfer learning after pretreatment: candidate path set generation algorithm (CTSG), similarity computing algorithm (SC), trajectory mapping algorithm (TM) and trigger duration transfer algorithm (TDT). A lot of experiments had been done in the end to verify the efficiency of activity pattern transfer in simulation environment. And the experiments present the methods good time consuming performance and effectiveness.
- Cloud computing,
- Conformal mapping,
- Distributed computer systems,
- Pattern recognition,
- Sustainable development,
- Trajectories,
- Activity patterns,
- Activity recognition,
- Similarity computing,
- Simulation environment,
- Smart environment,
- Trajectory transfer,
- Transfer learning,
- Trigger duration,
- Big data,
- Activity trajectory
Available at: http://works.bepress.com/sajal-das/68/