1. © 2005 Ritsumeikan Univ. All Rights Reserved. Context Aware Operation Reproduction for Safety Driving Satoshi Kaede Ritsumeikan University Graduate School of Computer Science Date Engineering Laboratory Japan E-mail kaede@de.is.ritsumei.ac.jp

2. © 2005 Ritsumeikan Univ. All Rights Reserved. Contents 1. Goal of Our Research 2. Describing Model of a Context 3. Verification of the Model 4. Conclusion and Future works

3. © 2005 Ritsumeikan Univ. All Rights Reserved. Goal of Our Research Set a Steering Lock ! Get Your Valuables ! We propose a method to reproduce operations from contexts of the driver and someone on a vehicle.

4. © 2005 Ritsumeikan Univ. All Rights Reserved. The method of representing human behavior The human behavior is consisted by individual act. Starting to drive a vehicle Open the door Get a key case Turn the key in the ignition Unlock the side brake Push the accelerator down

5. © 2005 Ritsumeikan Univ. All Rights Reserved. Bayesian Network The Bayesian Network It models dependency relation using probability networks. The structure of the Network is Directed Acyclic Graph.

6. © 2005 Ritsumeikan Univ. All Rights Reserved. Behavioral Scene Characteristic S1: A set of objects which are accessed when a user is taking a particular behavior. S2: A set of signals from ubiquitous environment when a user is taking the behavior. S3: A set of accessed objects and a set of signals from ubiquitous environment when a user is taking behaviors other than the one. S3 S1 S2 Behavioral Scene Characteristic

7. © 2005 Ritsumeikan Univ. All Rights Reserved. Bayesian Network The Bayesian Network It models dependency relation using probability networks. The structure of the Network is Directed Acyclic Graph. K2 Algorithm It automatically configures Bayesian Network by statistical data. It creates a dual directional arrow which represents dependency relationship between nodes. The allow interferes with configuration of DAG.

8. © 2005 Ritsumeikan Univ. All Rights Reserved. Result Switch to open a Gas Tank Lock Cap of Gas Tank Steering Lock Side Brake Switch of Automatic window Key Knob The Bayesian Network of Human Behavior

9. © 2005 Ritsumeikan Univ. All Rights Reserved. Heuristics Heuristic 1 An aim node is excluded from the set of candidate nodes which have a possibility becoming the parent node of all another nodes. Heuristic 2 2.1To cut the arrow that does not influence the aim node. 2.2To cut the arrow using semantics which the nodes have.

10. © 2005 Ritsumeikan Univ. All Rights Reserved. Result Switch to open a Gas Tank Lock Cap of Gas Tank Steering Lock Side Brake *1 *2 Switch of Automatic window Key Knob The Bayesian Network of Human Behavior

11. © 2005 Ritsumeikan Univ. All Rights Reserved. Describing Model of Human Behavior Heuristic 1 An aim node is excluded from the set of candidate nodes which have a possibility becoming the parent node of all another nodes. Heuristic 2 2.1To cut the arrow that does not influence the aim node. 2.2To cut the arrow using semantics which the nodes have. Heuristic 3 To sort the configured BNs which Heuristic 1, 2.1 and 2.2 are applied in ascending order using true cases and false cases.

12. © 2005 Ritsumeikan Univ. All Rights Reserved. RFID tag Access Log Accesses to RFID Tag Experiment for Verification of Model Database Server PDA with RFID reader Scenario : If a user leaves valuables in a vehicle when he leaves from vehicle for a long time, a system notifies him that valuables may be stolen by ruining on the vehicle.

13. © 2005 Ritsumeikan Univ. All Rights Reserved. True Cases False Cases dp 1 dp 2 Probability of leaving from vehicle 0 10 20 30 40 50 60 70 80 90 100 05101520253035 Case Analysis of experimental result

14. © 2005 Ritsumeikan Univ. All Rights Reserved. Conclusion and Future works We proposed the heuristics to configure behavioral scene characteristic from the context using Bayesian Network and the K2 Algorithm. Dual directional arrows are cut to configure candidate set of Bayesian networks by using proposed heuristics. We will get more experimental logs for verification model.

15. © 2005 Ritsumeikan Univ. All Rights Reserved. Thank you for listening. Thank you very much.

16. © 2005 Ritsumeikan Univ. All Rights Reserved. The Layer of Inferring Method The first stage: The BSC created from user contexts is checked with an access log and signals from ubiquitous environment. The check picks up behavior which may be occurring. The second stage: The behaviors which are picked up at the first stage are scrutinized as for the sequence of accesses and durations of accesses in access log, to determine whether the behaviors are really taken.

17. © 2005 Ritsumeikan Univ. All Rights Reserved. Tagged World Project Kitchen Bed Bathroom TV PC Table closet RFID Reader Access Log Access of RFID Tag Vestibule RFID Tags

18. © 2005 Ritsumeikan Univ. All Rights Reserved. The human behavior is consisted by individual act. Drive a vehicle Open the door Get a key case Turn the key in the ignition Unlock the side brake Push the accelerator down The Bayesian Network It models dependency relation using probability networks. The construct is Directed Acyclic Graph. K2 Algorithm It make construct of Bayesian Network by statistical data automatically. It creates a dual directional arrow which represents dependency relationship between nodes. The method of representing human behavior

19. © 2005 Ritsumeikan Univ. All Rights Reserved. Purpose of Our Research May I turn off an air conditioner ? May I turn off sound volume of stereo ? We propose a method to reproduce operations from contexts of the driver and someone on a vehicle.