The indoor positioning technology in this paper identifies real-time positions of the crew in ships. It can be used to get the regional personnel distribution and find timely locations of accidents. Therefore, the accuracy is very important when estimating positions in ships. This research considers the nearest reference point and beacon in ships as key points for improving the accuracy. We look for three key points to predict a user position. Firstly, KNN algorithm is used to get the first point P1. Secondly, the nearest reference point to the nearest beacon is the second point P2. Finally, the weighted centroid point of nearest beacons is the third point P3. The centroid position of these three key points is the predicted user position. Experimental results show the accuracy has been improved by at most 54%.