跨設備之Wi-Fi定位與地圖建構

Cross-Device Wi-Fi Localization and Mapping

顏孝杰(Hsiao-Chieh Yen)

指導教授 ： 周承復

共同指導教授 ： 王傑智(Chieh-Chih Wang)

國立臺灣大學/電機資訊學院/資訊工程學研究所/博士(2017年)

https://doi.org/10.6342/NTU201703979

摘要

本論文旨在改善在跨設備的場景下，以接收訊號強度(Received Signal Strength; RSS)作為量測訊號的Wi-Fi定位和同步定位與地圖建構(Simultaneous Localization and Mapping; SLAM)之精準度。跨設備之Wi-Fi定位面臨的主要挑戰在於每個設備間的RSS量測值有所差異。過往研究多使用一次線性方程式描述此差異。然而，該方程式並不能完整詮釋設備間的差異。因此，本論文提出使用線性回歸之殘差量化一Wi-Fi地圖經由線性跨設備轉換後所新增的不確定性。實驗發現該殘差與跨設備定位之誤差間具高度相關性。本論文因此提出一整合來自不同設備之地圖的方法，依據跨設備回歸殘差調整個別地圖的權重，以改善該組地圖對於任意新設備的定位精度。

由於Wi-Fi地圖之建構過程曠日廢時，過往研究提出Wi-Fi SLAM的方法企圖省去人工標註量測訊號正確位置的步驟。然而由於現有之Wi-Fi SLAM方法需對一非線性最佳化問題求解，其結果成敗取決於起始估測的好壞。由於Wi-Fi RSS不具備角度資訊，而未知角度是SLAM目標函數的一個主要非線性來源，本論文提出使用訊號強度之空間梯度(Signal Strength Gradient; SSG)推估Wi-Fi設備操作者的行走軌跡線段間之角度關係。經由推導與模擬實驗證實，兩相交線段間的角度差可由其SSG間的餘玹相關性近似，藉此即可由Wi-Fi RSS取得角度估測。實驗證實使用該角度估測進行純角度同步定位與地圖建構(Bearings-Only SLAM)可快速取得對整個地圖的精確角度起始估測。

關鍵字

Wi-FI定位 ； 設備多樣性 ； Wi-Fi同步定位與地圖建構 ； 方向性 ； 強健式同步定位與地圖建構

並列摘要

This thesis aims at improving the accuracy and reliability of cross-device Wi-Fi localization and Wi-Fi simultaneous localization and mapping (SLAM) using received signal strength (RSS) measurements. The main challenge in cross-device Wi-Fi localization, where different devices are used during training and testing, is that RSS measurements are device-specific. An extensive body of existing work has been using a linear function to map measurements between devices. However, past research has also found that such linear model is only a crude approximation, as RSS measurements are affected by many complex factors that are difficult to model. Instead of attempting to model each and every one of such factors, a dissimilarity measure is proposed to quantify the differences between devices unexplained by the linear mapping using the regression misfit. Experiments show that the amount of regression misfit is strongly related to cross-device localization performance and as such, the dissimilarity measure is used to weight maps from different devices in cross-device map fusion to improve localization accuracy.

As building any Wi-Fi map involves a labor-intensive training phase to provide location labels for all RSS measurements, Wi-Fi SLAM has been proposed in the literature to automate the process. However, existing Wi-Fi SLAM approaches involves optimizing a nonlinear cost function, which requires good initialization to approach a globally optimal solution. Finding a good solution is made more difficult because Wi-Fi RSS measurements does not provide orientation information, which is a major source of nonlinearity in SLAM problems. The thesis thus proposes an orientation constraint for Wi-Fi SLAM using the signal strength gradient (SSG) over trajectory segments. The cosine similarity between the SSGs over a pair of nearby segments is shown to approximate their relative orientations. This finding is used to create orientation constraints between near-parallel trajectory segments, transforming Wi-Fi SLAM into a bearing-only SLAM problem. Experiments show that the approach provides a fast and accurate orientation initialization for Wi-Fi SLAM. As SSGs are invariant to linear transforms in RSS space, the orientation constraints could also be used for cross-device Wi-Fi SLAM.

並列關鍵字

Wi-FI Localization ； Device Diversity ； Wi-Fi SLAM ； Orientation ； Robust Pose-Graph SLAM

參考文獻

Bahl, P. & Padmanabhan, V. N. (2000). Radar: An in-building rf-based user location and tracking system. In IEEE International Conference on Computer Communications (INFOCOM), (pp. 775–784).

Carlone, L., Aragues, R., Castellanos, J. A., & Bona, B. (2014). A fast and accurate approximation for planar pose graph optimization. International Journal of Robotics Research, 33(7), 965–987.

Chen, C., Chen, Y., Han, Y., Lai, H., Zhang, F., & Liu, K. J. R. (2017). Achieving centimeteraccuracy indoor localization on wifi platforms: A multi-antenna approach. IEEE Internet of Things Journal, 4(1), 122–134.

Chen, L.-H., Wu, E.-K., Jin, M.-H., & Chen, G.-H. (2014). Homogeneous features utilization to address the device heterogeneity problem in fingerprint localization. IEEE Sensors Journal, 14(4), 998–1005.

Cheng, H., y. Luo, H., & Zhao, F. (2012). Device-clustering algorithm in crowdsourcingbased localization. Journal of China Universities of Posts and Telecommunications, 19, 114–121.