No abstract

No abstract

No abstract

Scan registration plays an important role in robotics. The problem is to figure out the optimal transformation between two frames of data points, reference points and scene points. One approach to perform this task is using Normal Distribution Transform (NDT). The concept of NDT is to convert reference point cloud into Gaussian mixture distribution and then match scene points with this distribution. Nevertheless, this approach is likely to suffer from local optimum problem especially when the initial transformation error is large. Therefore, in recent researches, multi-scale registration technique is applied. In this work, multi-scale registration can be achieved by using polar scan clustering with different degree of separations. In registration problem, polar scan clustering has advantage over other clustering approaches because it is aware of the fact that the structure of data obtained from laser range finder is sorted by measured angle. To evaluate the performance of our approach, the experiment is conducted to compare transformation error resulted from our approach and a recent approach. As a result, the result from the proposed approach is better than the recent one significantly.