An Integrated Hierarchical Approach for Real-Time Mapping with Gaussian Mixture Model

To achieve effective collaboration of multiple robots, it requires efficient exchanges of map information. As directly exchanging generally used depth map requires high commu- nication bandwidth, it is practical to enhance the efficiency using map compression techniques based on Gaussian mixture models. Currently, parameters of the Gaussian mixture model are mostly computed using the expectation-maximization algorithm. It is time consuming as it has to iteratively update parameters by traversing all points in a point cloud converted from the depth map, and it is not suitable for real-time applications. Other methods directly segment the point cloud into grids and then perform a single Gaussian parameter estimation for each grid. They achieve real-time compression but generate parameter sensitive results. To tackle issues above, we improve compression methods with an integrated hierarchical approach. First, the points are clustered hierarchically and efficiently by K- means, generating coarse clusters. Then, each cluster is further hierarchically clustered by expectation-maximization algorithm for accuracy enhancement. After each clustering process, an evaluation index for ensuring accuracy and preventing over- fitting is calculated to determine whether pruning or retention of newly generated clusters is appropriate. At last, parameters of each Gaussian distribution in the model are estimated by points in a corresponding cluster. Experiments conducted in various environments demonstrate that our approach improves computing efficiency by over 79 times compared to the state-of- the-art approach.