An Energy Efficient Smartphone Sensors’ Data Fusion for High Rate Position Sampling Demands

Abstract

Many smartphone-based traffic safety applications have been proposed in literatures. These applications demand very high position sampling to safeguard vulnerable road users. European Telecommunication Standards Institute (ETSI) has defined time interval between Cooperative Awareness Messages for collision risk warning to be between Is and 0. 1\s. This implies that geographical awareness information has to be sampled between the frequencies 1Hz and 10Hz inclusive. However, an investigation we made depicts that current smartphones can’t support such high rate location sampling. Even though they meet the aforementioned sampling requirements, high rate sampling of position data is an energy hungry process. In light of this, we have proposed an energy efficient position prediction method that fuses GPS and Inertial Navigation Systems (INS) sensors data to estimate pedestrians’ positions at high rate. INS based dead reckoning is performed to extrapolate positions from last known location when GPS reading is unavailable and when GPS fix is realized the reading is used to correct dead reckoning parameters in addition to serving as a location fix. The proposed solution is compared to a position prediction method that relies solely on GPS data on two selected pedestrian trajectories. The result demonstrates that fusing GPS and INS position data has an average improvement of 30% and 61.4% in error in distance and direction respectively. The proposed position prediction algorithm is also applied to sensors data that are obtained by relaxing sampling rates with the objective of sparing smartphone’s energy. In this regard, first energy efficiency of different position sampling rates of GPS and INS sensors are evaluated and then the algorithm is applied to the sampling frequencies that are proven to husband energy. The outcome of the evaluation is that the battery life of smartphones can be doubled by compromising accuracy of estimated distance and direction by only 11.5% on average.