隨著定位技術的進步，使用無線定位技術能夠快速地定位人類。然而，位於室內環境的無線訊號傳輸，與能量衰減以及干擾息息相關，導致定位不穩定。最近，Google釋出使用Single Shot MultiBox Detector(SSD)模型建構的Tensorflow Object Detection API，可用於偵測人類。在本研究中，我們使用Tensorflow Object Detection API來進行偵測人物，並使用逆透視投影演算法(Inverse Perspective Mapping，IPM)精確地將其影像座標轉換為真實世界座標，透過降低影像序列幀率來準確與快速地進行偵測，並且定位人物。本研究提出的室內定位方法能夠以每秒一次準確地偵測人物，並將定位精準度提升到100厘米內。

With the advance of localization technology, wireless localization methods can position humans rapidly. However, the attenuation of signal transmission in the indoor environment caused by the Path loss and the noise leads to the instability of localizatioon. Proposed by Google recently, Tensorflow Object Detection API, which is structured by a SSD neural network model, can detect people accurately. In this paper, we apply the Tensorflow Object Detection API to detect humans, and transform their images coordinates to real world positions with Inverse Perspective Mapping (IPM). We can track humans in camera sequences accurately and rapidly localize them at the cost of lowering the frame rate. Our experimental results show that our proposed method can correctly detect humans once a second, and the accuracy of positioning then can be improved by less than 100cm.