基於影像檢索及行動感應器之室內導航系統

Abstract

本論文提出了一個基於影像檢索及行動感應器之室內導航系統，相對於現有的室內導航系統需要額外的硬體基礎設施，本系統僅須利用手機上的行動感應器和相機即可實現。 MPEG CDVS提供本系統影像檢索的基礎，根據圖像搜尋的結果判斷相對應的室內位置。但是大多數的室內環境存在重複的結構(例如走廊)和沒有影像特徵的結構(例如白牆)，這些潛在因素對影像檢索的結果造成負面的影響。為了能解決這些議題，本系統提出了一個結合手機行動感應器和影像檢索的機率模型來提升室內導航系統的準確度。此外，本系統亦設計了一個基於信息量的影像過濾器，藉此過濾掉信息量不足的影像輸入。 實驗證實本系統能做為一室內導航的基礎。這更是MPEG CDVS架構完整移植到安卓手機平台上的首例，除了將影像檢索的程式完整的在手機上呈現外，也展示了基於影像檢索之室內導航系統未來標準化的可能性。

The standard of MPEG-7 Compact Descriptor for Visual Search (CDVS) framework is the foundation of the visual search component. However, the repeating structures throughout the indoor environment (e.g. corridors) and locations with little visual features (e.g. uniform walls) can potentially lead to low accuracy of the vision-based system. To cope with this issue, a probabilistic model is proposed to fuse the location associated image retrieval result by CDVS and the sensors readings on mobile device. Furthermore, an entropy-based fi?lter is designed to fi?lter those input images without su?fficient information gain. Experiments show good results for the proposed system. Moreover, it is the ?rst time that the complete MPEG-7 CDVS framework is implemented on mobile Android platform, which not only ensures the proposed system can work in offl?ine mode, but also demonstrates the future possibility on vision-based indoor navigation system to standardize.