I have developed and evaluated a data reduction system for point data by using rendering the target 3D object represented with points from multi directions. So far, although this method has been applied to surface data, it was extended to point data in this study. For point data, this method works as a filter that extracts points with sufficient density from the original points. In this case, it is possible to solve the duplication extraction problem which occurs with surface data, by using a point ID buffer that works similar to the Z buffer. I have implemented the prototype system with the function and evaluated the performance. In addition, the characteristics of the two parameters required for the proposal method was investigated. I confirmed that the system works effectively and learned that the total number of rendered pixels determined by the two parameters has a positive correlation with the quality of the extracted point. Occlusion at rendering is a problem for complicated shapes. In such cases, it was suggested that, if the total rendering pixel numbers are same, increasing the number of the rendering direction can improve the quality more than increasing the resolution per a rendering image.