自導航技術誕生以來，經過不斷的演變與發展，已成為現代高科技城市不可或缺的領域，且廣泛使用於車載、船隻、航空、行人等不同面向的應用。然而，現今最成熟的室外定位技術即為全球導航衛星系統(Global Navigation Satellite System, GNSS)與慣性導航系統(Inertial Navigation System, INS)。全球導航衛星系統卻大的缺點即為易受訊號遮蔽導致定位精度大幅受限，像是在城市峽谷的多路徑效應或是隧道內的訊號中斷等都是經典的衛星定位受限場景。慣性導航系統的缺點則是慣性測量元件(Inertial Measurement Unit, IMU)的感測器誤差會隨著時間增加而累積成飄移誤差。因此，慣性導航系統與全球導航衛星系統的整合為現今室外導航最被廣泛使用的演算法。本研究所探討的行人導航之原理也是基於這兩種系統的整合為基底去做進階的延伸。
近年來，穿戴式感測器及行動裝置成為行人導航及定位領域上的熱門工具。在行人導航領域中，行人航向推算(Pedestrian Dead Reckoning, PDR)為最主要之演算法。其主要利用慣性測量元件(後面皆以IMU稱之)提供之加速度資訊來偵測步伐和推算人體步長，以及旋轉量資訊來推估航向，經積分整合得出位置變化。然而傳統行人航向推算(後面皆以PDR稱之)之步長計算經驗公式受使用者身高、走路頻率、走路習慣等因素影響，若沒有設定合適的公式參數，將會導致步長估算有很大的誤差，進而導致定位成果不佳。除了步長計算的問題之外，IMU本身就含有系統誤差及噪聲，會隨時間產生飄移誤差，此問題在消費者等級的IMU上更為顯著。
由於在行人應用上，手機可能會出現在不同位置(傳訊息、講電話、握在手上…等等)，本研究先建立了長短期記憶模型(Long Short-Term Memory, LSTM)來預測手機當下時刻的位置，再使用卷積神經網路(Convolutional Neural Network, CNN)來創建速度估計模型，用以估算使用者的一維速度，藉由訓練不同使用者的數據，以取得比傳統步長計算更穩定的位移量。航向部分本研究使用姿態估算系統(Attitude and Heading Reference System, AHRS)中的積分法取得預估航向，再以最新型9D IMU AHRS演算法的高精度航向解作為觀測更新，近一步以擴增卡曼濾波器(Extended Kalman Filter, EKF)之原理進行航向融合以取得整合解。最後加入全球導航衛星系統並同樣基於擴增卡曼濾波器之原理去做兩種感測器之資訊與系統整合，以此彌補IMU隨時間產生飄移之問題。
本研究之實驗配置，參考軌跡使用NovAtel Pwrpak E2，其內件高規格GNSS及IMU可產出精度良好之整合解，可視為地真參考。並使用Huawei Mate 20 pro做為行動裝置測試儀器與其進行不同模式之軌跡比較。實驗場域安排全透空以及半透空(途中經過高樓間或樹蔭下)，以證實此系統在衛星訊號不良時依然能依靠速度估計模型與航向推算而擁有穩定可靠的定位成果。

Navigation technology has continuously developed and become indispensable in modern high-tech cities. It is widely used in various applications such as vehicle, ship, aviation and pedestrian navigation. However, the most mature outdoor positioning technologies nowadays are the Global Navigation Satellite System (GNSS) and the Inertial Navigation System (INS). The major disadvantage of GNSS is that the signal is easily blocked, which can significantly limit positioning accuracy. Classic scenarios where satellite positioning is restricted include multipath effects in urban canyons or signal outage in tunnels. The main disadvantage of INS is that the sensor errors from the Inertial Measurement Unit (IMU) accumulate over time, resulting in drift errors. Therefore, the integration of INS and GNSS is currently the most commonly used algorithm for outdoor navigation. The principles explored in this study for pedestrian navigation are also based on the fusion of these two systems, with further advanced extensions.
In recent years, wearable and mobile devices have become popular tools in pedestrian navigation. In pedestrian navigation, Pedestrian Dead Reckoning (PDR) is the main algorithm. It utilizes acceleration provided by the IMU to detect steps and estimate step length, and rotational information to estimate heading to derive position changes. However, traditional step length formula is influenced by factors such as the user's height, walking frequency, and walking habits. With inappropriate parameters set, it can lead to significant errors in step length estimation and thus poor positioning results. Besides, IMU itself contains system errors and noise, leading to drift errors over time, which is more serious in consumer-grade IMUs.
Since mobile device could be possessed in different modes (e.g., texting, calling, swinging, etc.), this study first established a Long Short-Term Memory (LSTM) model to predict the phone's location at any given moment. Then, a Convolutional Neural Network (CNN) is used to create a velocity estimation model to estimate the user's 1D velocity. By training with data from different users, the study aims to achieve more stable displacement results compared to traditional step length calculations. For heading estimation, we use the integration method in the Attitude and Heading Reference System (AHRS) to obtain an estimated heading, then uses the high-precision heading solution of the latest 9D IMU AHRS algorithm as an measurement update. Extended Kalman Filter (EKF) algorithm is further used for heading fusion solution. Finally, GNSS is combine in the system also based on EKF to compensate for the drift issue of the IMU over time.
For the experimental setup in this study, NovAtel Pwrpak E2 is referred to as reference trajectory, since it contains high-grade GNSS and IMU capable of producing high-precision fusion solutions. Huawei Mate 20 pro is used as the testing mobile device, and its trajectory is compared in different modes. The experimental field includes both fully open and semi-open sky (e.g., passing between tall buildings or under tree shades) to demonstrate that this system can still rely on the speed estimation model and heading calculation to achieve stable and reliable positioning results when satellite signals are poor.

[1] Asraf, O., Shama, F., & Klein, I. (2022). PDRNet: A Deep-Learning Pedestrian Dead Reckoning Framework. IEEE Sensors Journal, 22(6), 4932–4939. https://doi.org/10.1109/JSEN.2021.3066840
[2] Capital Partners, W. (2016). Beyond The Headlights: ADAS and Autonomous Sensing Eye/Face Tracking. www.woodsidecap.com
[3] Chen, C., Lu, X., Markham, A., & Trigoni, N. (n.d.). IONet: Learning to Cure the Curse of Drift in Inertial Odometry. www.aaai.org
[4] Deng, Z. A., Wang, G., Hu, Y., & Wu, D. (2015). Heading estimation for indoor pedestrian navigation using a smartphone in the pocket. Sensors (Switzerland), 15(9), 21518–21536. https://doi.org/10.3390/s150921518
[5] Diaz, E. M., de Ponte Muller, F., Jimenez, A. R., & Zampella, F. (2015). Evaluation of AHRS algorithms for inertial personal localization in industrial environments. Proceedings of the IEEE International Conference on Industrial Technology, 2015-June(June), 3412–3417. https://doi.org/10.1109/ICIT.2015.7125605
[6] El-Sheimy, N., & Li, Y. (2021). Indoor navigation: state of the art and future trends. In Satellite Navigation (Vol. 2, Issue 1). Springer. https://doi.org/10.1186/s43020-021-00041-3
[7] Falco, G., Pini, M., & Marucco, G. (2017). Loose and tight GNSS/INS integrations: Comparison of performance assessed in real Urban scenarios. Sensors (Switzerland), 17(2). https://doi.org/10.3390/s17020255
[8] Fang, J., Sun, H., Cao, J., Zhang, X., & Tao, Y. (2011). A novel calibration method of magnetic compass based on ellipsoid fitting. IEEE Transactions on Instrumentation and Measurement, 60(6), 2053–2061. https://doi.org/10.1109/TIM.2011.2115330
[9] Global IT Research Institute, IEEE Communications Society, & Institute of Electrical and Electronics Engineers. (n.d.). The IEEE 17th International Conference on Advanced Communication Technology : “Named Data Networking - A Future Internet Architecture” : ICACT 2015 : Phoenix Park, PyeonhChang, Republic of Korea : Jul. 1-3, 2015 : proceeding.
[10] Guo, Q., Jin, S., Li, M., Yang, Q., Xu, K., Ju, Y., Zhang, J., Xuan, J., Liu, J., Su, Y., Xu, Q., & Liu, Y. (2020). Application of deep learning in ecological resource research: Theories, methods, and challenges. In Science China Earth Sciences (Vol. 63, Issue 10, pp. 1457–1474). Science in China Press. https://doi.org/10.1007/s11430-019-9584-9
[11] Xu, L., Xiong, Z., Liu, J., Wang, Z., & Ding, Y. (2019). A novel pedestrian dead reckoning algorithm for multi-mode recognition based on smartphones. Remote Sensing, 11(3). https://doi.org/10.3390/rs11030294
[12] Hou, X., & Bergmann, J. (2021). Pedestrian Dead Reckoning with Wearable Sensors: A Systematic Review. In IEEE Sensors Journal (Vol. 21, Issue 1, pp. 143–152). Institute of Electrical and Electronics Engineers Inc. https://doi.org/10.1109/JSEN.2020.3014955
[13] Institute of Electrical and Electronics Engineers. (n.d.). 2020 IEEE International Conference on Robotics and Automation (ICRA) : 31 May-31 August, 2020. Paris, France.
[14] Jiang, X., Hadid, A., Pang, Y., Granger, E., & Feng, X. (2019). Deep learning in object detection and recognition. In Deep Learning in Object Detection and Recognition. Springer Singapore. https://doi.org/10.1007/978-981-10-5152-4
[15] Kang, J., Lee, J., & Eom, D. S. (2018). Smartphone-based traveled distance estimation using individual walking patterns for indoor localization. Sensors (Switzerland), 18(9). https://doi.org/10.3390/s18093149
[16] Karimi, Z., Soheili, M., Heydarishahreza, N., Ebadollahi, S., & Gill, B. (2021). Smartphone Mode Detection for Positioning using Inertial Sensor. 2021 IEEE 12th Annual Information Technology, Electronics and Mobile Communication Conference, IEMCON 2021, 209–215. https://doi.org/10.1109/IEMCON53756.2021.9623130
[17] Kim, J. W., Jang, H. J., Hwang, D.-H., & Park, C. (2004). A Step, Stride and Heading Determination for the Pedestrian Navigation System. Journal of Global Positioning Systems, 3(1 & 2), 273–279. https://doi.org/10.5081/jgps.3.1.273
[18] Klein, I. (2020). Smartphone location recognition: A deep learning-based approach. Sensors (Switzerland), 20(1). https://doi.org/10.3390/s20010214
[19] HEXAGON. (2023). An Introduction to GNSS.
[20] Klein, I., & Asraf, O. (2020). StepNet - Deep Learning Approaches for Step Length Estimation. IEEE Access, 8, 85706–85713. https://doi.org/10.1109/ACCESS.2020.2993534
[21] Klein, I., Solaz, Y., & Ohayon, G. (2018). Pedestrian Dead Reckoning with Smartphone Mode Recognition. IEEE Sensors Journal, 18(18), 7577–7584. https://doi.org/10.1109/JSEN.2018.2861395
[22] Krstinić, D., Braović, M., Šerić, L., & Božić-Štulić, D. (2020). Multi-label Classifier Performance Evaluation with Confusion Matrix. 01–14. https://doi.org/10.5121/csit.2020.100801
[23] Kshetrimayum, R. S. (2009). An introduction to UWB communication systems. IEEE Potentials, 28(2), 9–13. https://doi.org/10.1109/MPOT.2009.931847
[24] Kuang, J., Li, T., & Niu, X. (2022). Magnetometer Bias Insensitive Magnetic Field Matching Based on Pedestrian Dead Reckoning for Smartphone Indoor Positioning. IEEE Sensors Journal, 22(6), 4790–4799. https://doi.org/10.1109/JSEN.2021.3073397
[25] Kuang, J., Niu, X., Zhang, P., & Chen, X. (2018). Indoor positioning based on pedestrian dead reckoning and magnetic field matching for smartphones. Sensors (Switzerland), 18(12). https://doi.org/10.3390/s18124142
[26] Laidig, D., & Seel, T. (2022). VQF: Highly Accurate IMU Orientation Estimation with Bias Estimation and Magnetic Disturbance Rejection. https://doi.org/10.1016/j.inffus.2022.10.014
[27] Maruyama, Y., Kato, S., & Azumi, T. (2016, October 1). Exploring the performance of ROS2. Proceedings of the 13th International Conference on Embedded Software, EMSOFT 2016. https://doi.org/10.1145/2968478.2968502
[28] Noureldin, A., Karamat, T. B., & Georgy, J. (2013). Fundamentals of inertial navigation, satellite-based positioning and their integration. In Fundamentals of Inertial Navigation, Satellite-Based Positioning and their Integration. Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-30466-8
[29] Thesis, H., & Mautz, R. (n.d.). ETH Library Indoor positioning technologies. https://doi.org/10.3929/ethz-a-007313554
[30] Vandermeeren, S., & Steendam, H. (2022). Deep-Learning-Based Step Detection and Step Length Estimation with a Handheld IMU. IEEE Sensors Journal, 22(24), 24205–24221. https://doi.org/10.1109/JSEN.2022.3219412
[31] Weinberg, H. (n.d.). Using the ADXL202 in Pedometer and Personal Navigation Applications. www.analog.com/library/applicationNotes/
[32] Yan, H., Herath, S., & Furukawa, Y. (2019). RoNIN: Robust Neural Inertial Navigation in the Wild: Benchmark, Evaluations, and New Methods. http://arxiv.org/abs/1905.12853
[33] Yan, H., Shan, Q., & Furukawa, Y. (2017). RIDI: Robust IMU Double Integration. http://arxiv.org/abs/1712.09004
[34] Yousif, K., Bab-Hadiashar, A., & Hoseinnezhad, R. (2015). An Overview to Visual Odometry and Visual SLAM: Applications to Mobile Robotics. Intelligent Industrial Systems, 1(4), 289–311. https://doi.org/10.1007/s40903-015-0032-7
[35] Zhang, P., Chen, X., Ma, X., Wu, Y., Jiang, H., Fang, D., Tang, Z., & Ma, Y. (2017). SmartMTra: Robust indoor trajectory tracing using smartphones. IEEE Sensors Journal, 17(12), 3613–3624. https://doi.org/10.1109/JSEN.2017.2692263
[36] Zhang, R., Mi, J., Li, J., & Wang, Q. (2022a). A Continuous PDR and GNSS Fusing Algorithm for Smartphone Positioning. Remote Sensing, 14(20). https://doi.org/10.3390/rs14205171
[37] Zhang, R., Mi, J., Li, J., & Wang, Q. (2022b). A Continuous PDR and GNSS Fusing Algorithm for Smartphone Positioning. Remote Sensing, 14(20). https://doi.org/10.3390/rs14205171
[38] Li, M., Chen, R., Liao, X., Guo, B., Zhang, W., & Guo, G. (2020). A precise indoor visual positioning approach using a built image feature database and single user image from smartphone cameras. Remote Sensing, 12(5). https://doi.org/10.3390/rs12050869