We develop the context information prediction system using multi-sensor and indoor localization technologies based on fingerprinting technique as shown in Fig. 1, which is divided into learning, estimation, and prediction phases. At learning phase, the training terminal on the mobile robot sends signals, its location, and context information to APs. The APs convert the received signals into the propagation parameters (RSSI and RSPD) and then transfer them to AP controller (APC). APC receives and stores them in the database. Application server (AS) downloads them from the APC database and constructs radio and context map by the information. At estimation phase, a target terminal sends signals and walking information to APs. As with the same as the learning phase, AS receives the propagation parameters and walking information from the APs. Then, AS estimates the location of the target terminal by pattern matching with the parameters and radio map and walking information. At prediction phase, AS predicts the location of the target terminal ahead by N terms based on the estimated location and walking information; and then context information with context map and predicted location. If the predicted context information exceeds a threshold, AS warns target terminal.
Figure. 1 Developed context information prediction system
We also evaluate the localization performance by an indoor experiment and the context information prediction performance on the simulator. The experimental parameters, environment and results are shown in Tab.1, Fig. 2 and 3 respectively.
Table 1. Experimental parameters.
In the experiment, the mobile robot moves along the movement path in learning phase and the target terminal moves along the walking route in estimation phase. The results verify that RSPD+RSSI+walking information scheme achieves mean estimation error 1.36 m. In the simulator, AS predicts the context information using the location obtained from the above experiment and virtual context map. The results verify that false positive achieves 24% when probability of miss detection is 5% and N is 2 in Fig. 4.
Figure 2. Experimental environment.
Figure 3. Location Estimation
Figure 4. Context prediction Error. accuracy.
- R. Kosaka, G.K. Tran, K. Sakaguchi, K. Araki, “[Poster Presentation] Indoor Localization System with Wi-Fi Fingerprints and Pedestrian Dead Reckoning ,” CQ Technical Committee, IEICE technical report, Vol. 2016-10-CQ, Oct. 2016.