Robust algorithms for fingertips detection

This study developed fingertips detection algorithms for contactless applications of human-machine interaction. The proposed methods are capable of capturing high-quality hand models without being affected by external influences, such as changes in lighting or complex backgrounds. The proposed system is based on the offline construction of databases comprising skin color and non-skin color samples. The means and standard deviations of these samples respectively serve as the initial cluster centers of the K-means classifier and the screening radii. In this way, we were able to accelerate the speed of convergence in the classification of clusters to enhance the accuracy of hand model images. The circular sweeping method and breadth-first search (BFS) algorithm are used to detect fingertips after the edges of hand image are established. The robustness of the proposed method was experimentally verified in terms of the skin recovery rate and the accuracy of fingertip detection, which remained effective despite variations in skin color and the complexity of the background.