BIR-AHC: Balanced Iterative Reducing and Agglomerative Hierarchical Clustering for Stair Detection

Recently, stair environment perception has attracted considerable attention in humanoid robots. In complex stair environments, there are some problems, such as incomplete stairs, obstacle occlusion, and unequal stair heights, resulting in relatively low accuracy in constructing stairs. To solve this problem, we propose a balanced iterative reduction and cohesive hierarchical clustering method for stair environment perception, which consists of two main parts: a point cloud preprocessing module and a stair modeling module. Specifically, given a fixed resolution point cloud, we first preprocess and initially cluster the data using a hierarchy-based balanced iterative reduction and clustering algorithm to reduce point cloud noise and establish contextual links between point clouds. Then, we construct the graph between points and calculate the mean square error between individual points on this basis to efficiently construct the stair plane and effectively improve the accuracy of the 3D model of the stair. To evaluate the effectiveness and efficiency of the algorithm, we conduct experiments on RGB-D sensor datasets collected from different scenes, and the experimental results show that the proposed algorithm has a more satisfactory performance in terms of modeling accuracy and computation time for the stair environment.