For the last decade multibeam sonars have been increasingly used for mapping and visualization of the seafloor to provide the ``physical bases'' for environmental studies. Increasing amount of digital (raster) echo records of high resolution from a multibeam sonar have enhanced the potential of computer modeling of the marine environment to improve our understanding of the bottom processes. However, the 3D bottom images as the result of merging different sonar transects do not comply exact geographical positions and should be corrected. Additionally, the raw sonar records are subject to systematic errors, random noise and outliers. In this paper, Kalman filtering techinque to generating optimal estimates of bottom surface from a noisy raw sonar records is proposed. The experiment on the surface indicates that after applying the Kalman filtering the outliers of raw records can be efficiently removed. Moreover, the two-step Kalman filtering method enables 3D seabed visualization in real time. The paper proposes the geographical corrections applied to the merged mutibeam sonar transects records. The 3D bottom relief before, and after the filtering method are presented.