Abstract: In an underwater simultaneous localization and mapping (SLAM) system, the autonomous underwater vehicle (AUV) needs to estimate its own pose while building a map of the environment. The pose of the AUV can be predicted by dead reckoning, but navigation errors accumulate over time. Therefore, additional observation means are required to correct the AUV pose. A multibeam echosounder (MBES) can acquire bathymetric point cloud data containing environmental depth information, and is one of the commonly used measurement devices in underwater SLAM. However, when the overlap rate of underwater terrain data is low, data association becomes considerably difficult. Existing studies have mostly focused on the iterative closest point (ICP) method and its variants, but when the overlap rate decreases, the processing performance of these methods on bathymetric data remains limited. To address the above problem, this paper proposes an improved segmented voting method, and conducts two groups of bathymetric submap registration experiments based on measured data. The experimental results show that, compared with commonly used methods, the proposed method can achieve point cloud registration more effectively under low overlap rates.