A performance model of pedestrian dead reckoning with activity-based location updates

Abstract

Advanced computing and sensing capabilities of smartphones provide new opportunities for personal indoor positioning. A particular trend is to employ human activity recognition for autonomous calibration of pedestrian dead reckoning systems thereby achieving accurate indoor positioning even in the absence of any positioning infrastructure. The basic idea is that the activity context, such as switching from a walking to a stair climbing activity gives clues about pedestrian's current position. In this paper, we have made a first attempt in developing a performance model for such systems. For an unbiased random walk, we have obtained two interesting results in closed-form expressions. First, we have demonstrated that the distance a pedestrian is expected to travel before the PDR is recalibrated is reciprocal of the density of activity switching points (ASPs) in the indoor environment. The implication of this finding is that the continuous unaided use of PDR can be curbed drastically by identifying more ASPs in a given environmental setting. Second, we have shown that false negatives of the activity detection algorithms do not have a major impact as long as they are within a reasonable range of 0–30%. The system performance however degrades rapidly if false negatives continue to grow beyond 30%.