Spatial Geometric Constraints based Iterative Clustering Algorithm

In visual measurement technology, the precise extraction of spatial feature points’ centroids is crucial to ensure the measurement accuracy of the visual system based on feature point imaging. This task is particularly challenging because of light spot blurring, gray influence, random shapes, gray impacting, and modeling limitations. The current extraction methods for light spot center can only extract single spot center at each time. As far as for multiple light spots in image, extracting their centers must be manually one by one. In this study, a spatial geometric constraints based on iterative clustering algorithm (SGCICA) is proposed and automatically extract multiple light spots’ centers through a K-means algorithm practice. Considering clustering algorithms can easily obtain multiple cluster centers in feature space, we introduce the information in image space into clustering algorithms from two aspects: (1) the pixel coordinate is adopted as the features for clustering algorithm to obtain the multiple light spots’ centers; (2) the spatial orders and geometric constraints among the spots are defined in the objective function of clustering algorithms to ensure the accurate extraction of actual LED targets. SGCICA operating in clustering feature space can effectively and naturally manage the information from image space. The spatial geometric constraints can improve the precision and the robustness of the clustering results. In experiments, the noise resistance and extraction precision of the proposed method are evaluated using synthetic and real data and compared with the existing light spot centers extraction methods and clustering algorithms. Both qualitative and quantitative measures indicate that the precision of the extracted light spot centers by SGCICA is kept within 0.04 pixels and satisfy the known order and spatial geometric constraints.