Abstract: Vision Localization has become a crucial problem to be solved with the continuous popularity of sweeping robots. However, due to the limited resources of embedded systems, it is difficult to be realized and we can hardly get good performance. To address this challenge, in this paper the monocular camera which faces the ceiling is used to reduce the computational complexity and solve the uncertainty of monocular scale. Firstly, we extract and match the feature points of collected consecutive frames by using the FAST with Shi-Tomasi algorithm and ORB matching algorithm, and jointly consider the vision localization model to estimate the global positions. Then, the ORB matching algorithm is improved using the rough data provided by inexpensive sensors. Finally, we design a UI of feature map to perform visual experiments and the results show that our proposed scheme has high accuracy and low computational complexity.