In this study, the methods of pedestrian dead reckoning were implemented on a handheld embedded system by using measuring raw data obtained from inertial measurement unit sensors (IMU) and the best method was proposed after comparing them. The location estimation algorithms for pedestrians respectively are the orientation estimation, the step detection and the adaptive step length estimation. The low errors of these algorithms will contribute to obtain better location estimation results with high accuracy. Therefore, the factors causing errors and noises on especially handheld systems were probed and elimination of these problems was worked on. Noises and measurement errors occurred during stepping were eliminated by using Kalman Filter. In addition, the errors due to hand deviation during stepping were also eliminated by proposing conditional orientation algorithm. Finally, the method that gives better result was proposed by combining these algorithms.