同時定位與地圖建置 (simultaneous localization and mapping, SLAM) 演算法，是利用移動中的光學感測器取得 3D 環境資訊的核心技術之一，廣泛應用在自動駕駛車、自走車(automated guided vehicles, AGV)、和家用機器等。有時候，SLAM 演算法的效益很大程度上依賴於光學設備，像是相機、LIDAR 等，取決於輸入的信號品質。而在 SLAM 演算法中的其中一個步驟為閉環檢測，這個步驟為的是要檢測機器人是否到達曾經到過的位置，進而消除在建立地圖時所累積的誤差，成為 SLAM 中的重要步驟。一般藉由幾何特徵描述來判斷照片裡的場景是否為相似，但是有時候在接近相同的場景時，幾何特徵的方法仍然沒有效。因此，我們可以將物件辨識以及時間－空間序列這兩個特徵去做比較，並把它們整合到 SLAM 的流程中，可以呈現更準確的 3D 空間資訊。在這篇文章當中，首先我們會提供SLAM 的概述，再來是介紹物件辨識和時間－空間序列的比較方法，這兩個方法為的是在SLAM 中更可以比較照片中所出現的場景是否相似，經由辨識物件像是地標或標誌甚至是物件，我們可以更好的分類相似的場景，並且改善室內 3D 建圖的結果。

Simultaneous localization and mapping (SLAM) algorithm is one of the core technologies to build 3D environment data using optical sensors. The technology is widely utilized on machines such as automated guided vehicles (AGV) and domestic robots. The performance of SLAM algorithms is highly depending on both the software and the quality of optical sensors, like cameras, LIDAR, etc. Loop closure, one of the crucial SLAM function component to highlight, is responsible for detecting visited locations and correcting accumulated errors. Conventionally, loop Closure calculates the similarity of scenes by comparing geometric features, but in scenarios where different scenes appears nearly identical, the performance of feature-based methods degrade significantly. Therefore, instead of using geometric feature, we introduce object recognition combining with time-spatial sequences to evaluate the similarity and improve the SLAM process. In this article, we first overview the SLAM process, and then give the introduction with object recognition and time-spatial comparing schemes. By identifying landmarks objects or signs, we can better classify similar scenes and improve 3D indoor mapping results.