There are two major approaches to implement the Pedestrian Dead Reckoning system, one of the most widely used indoor localization algorithm using an IMU: Integrated Approach (IA) adopts the conventional strapdown INS system and is frequently used in foot-mounted systems. For handheld devices, on the other hand, the Parametric Approach (PA) is selected more often since no measurement can be used to compensate for the rapidly increasing drift errors. The PA-based PDR system uses the information between the two successive steps to update the position of the device, making the robustness of the system rely heavily upon accurate step detection. The most frequently used algorithm, the accelerometer peak detection, however, suffers from a high miss detection rate from various disturbances including a normal arm swinging and a motion transition. In this paper, we propose a directed peak detection method to improve the accuracy of the step detection. The acceleration during step is generated as the foot touches the ground, and therefore is nominally directed vertically upwards. Since the attitude information is estimated continuously in the PA-based PDR system, the direction of the detected peak can be evaluated in the inertial frame. The proposed method was tested on PennCOSYVIO data set to verify accuracy improvement.

Keywords: indoor navigation, pedestrian dead reckoning, step detection