基於手持移動裝置之室內空間文字影像擷取

Text Images Based Spatial Information Retrieval with Handheld Mobile Devices

陳加容(Chia-Jung Chen)

指導教授 ： 黃乾綱

國立臺灣大學/工學院/工程科學及海洋工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201602757

摘要

近年來隨著科技的進步、無線感知設備與行動裝置普及，陸續發展眾多空間定位的應用服務。談到定位與導航的應用，人類現今在室外最常使用即是全球定位系統 (Global Positioning System, GPS)。人類在生活中，有很大一大部分的時間都從事室內活動，但由於室內環境的限制，使得 GPS 無法在室內空間中做精確定位服務，進而產生出室內定位服務的需求。

本論文利用人類在面對陌生公共環境，且 GPS 無法提供有效服務時，亦採取尋找方向告示牌、警示牌、路標或室內地圖…等文字說明訊息的行為，提出一演算法以模擬人類在陌生公共室內環境時，會採取的視覺影像資訊檢索暨定位策略做為導航之用。

本論文以行動設備的相機模擬人類視覺的感知設備，在室內空間中拍攝具有文字資訊的方向指示牌，接著透過影像處理最大穩定極值區域 (MSER) 的方法，偵測影像中具有資訊的區域，經過演算法的計算，擷取出影像中的文字資訊區域，再利用尺度不變特徵轉換 (SIFT) 匹配預先建立的室內文字影像地圖，以達到室內空間定位的目的。

本論文之演算法在未經過數據訓練的情況下，文字影像偵測正確率達到 74.84%，且影像是由行動設備在人潮眾多的台北車站捷運站內以及其它大眾運輸室內環境所拍攝。而在文字影像空間定位實驗中，室內定位的正確率達 79.89%。證實本論文之演算法在無建置室內空間模型、訓練數據與控制環境條件下，亦可以達到室內空間定位的目的。

關鍵字

室內定位 ； 文字影像偵測 ； 最大穩定極值區域 ； 尺度不變特徵轉換

並列摘要

Outdoor GPS is the backbone of positioning and navigation applications, today. However, people engage in indoor activities much more than outdoor activities in an urban environment. Unfortunately, GPS can not provide precise positioning service in an indoor setting. Therefore, the need for indoor positioning services emerges.

The approach proposed in this study mimics human’s natural behavior to find the directions from traffic signs, warning signs, indoor maps in an unfamiliar public environment without precise GPS service. In other words, this study takes the strategy of retrieving image and positioning information for the purpose of navigation in an unfamiliar public indoor environment.

This study takes mobile device's camera as human visual input and takes pictures of the direction signs with text information in indoor space. Then, a MSER-based feature detector is adopted to detect image regions with information as candidates. The algorithm extracts the images of the text information area with SIFT feature detection to match pre-established text image maps to serve the purpose of spatial information retrieval.

The algorithm proposed in this study does not require any training works before installation, therefore it saves training time and avoids the overfitting problem. The images were taken with a mobile device in Taipei MRT Station and other public transport indoor environment. The text detection experiment has 74.84% precision. Basing on the experiment, a further text-image spatial positioning experiment is conducted and reached 79.89% precision. According to the results, the proposed algorithm without building a spatial model and training data in advance is robust and applicable for uncontrolled environmental image inputs. In other words, it can be the core part of a service for an indoor positioning system in the future.

並列關鍵字

Indoor Localization ； Text Detection ； MSER ； SIFT

參考文獻

[37] Alcantarilla, Pablo Fernández, Adrien Bartoli, and Andrew J. Davison. "KAZE features." European Conference on Computer Vision. Springer Berlin Heidelberg, 2012.

[1] Pahlavan, Kaveh, Xinrong Li, and Juha-Pekka Makela. "Indoor geolocation science and technology." IEEE Communications Magazine 40.2 (2002): 112-118.

[2] Newman, Nic. "Apple iBeacon technology briefing." Journal of Direct, Data and Digital Marketing Practice 15.3 (2014): 222-225.

[4] Mori, Shunji, Ching Y. Suen, and Kazuhiko Yamamoto. "Historical review of OCR research and development." Proceedings of the IEEE 80.7 (1992): 1029-1058.

[6] K. Jung, K. I. Kim, and A. K. Jain. "Text information extraction in images and videos: A survey." Pattern Recognition, 37(5):977–997, 2004.