Abstract
In this thesis, we provide help to the visually impaired people navigation assistance by campus landmark recognition. By combining modern smart phone with the image matching techniques in computer vision, the user can obtain the target image on their smart phone, and transport the target image to the server by a wireless network. On the server side, we extract the Speeded-Up Robust Feature (SURF) features for matching against an image database. Finally, the user can receive the best matching result and recognize the target building. The database on the server side is constructed before runtime. In this research, we focus on developing a method of constructing an efficient database. The contribution of this thesis is to refine an appropriate amount of images with highly representative as template database from large-scale training data based on feature selection. The template database refined by our method has the capacity of maintaining the balance between accuracy and speed to achieve a high performance. To show the usefulness of our method, we have implemented the proposed method to run on Android based smart phones to help the visually impaired students to navigate the campus. Experimental results show that our refined database can recognize landmark more effectively comparing to baseline.