Multi-Feature Fusion Fall Detection Using OpenPose and Analytic Hierarchy Process

Human fall detection plays an important role in various fields such as medical monitoring, industrial safety, and sports analysis. This study proposes a fall detection method based on the OpenPose human pose estimation model, which utilizes human keypoints and skeletal information. An Analytic Hierarchy Process (AHP) is integrated to construct a multi-feature fusion framework. First, the OpenPose-MPI model is used to extract skeletal keypoints and generate the human bounding box. Then, both static features (aspect ratio of the bounding box) and dynamic features (left knee joint angle and centroid velocity) are combined to build a multi-criteria decision model. Finally, AHP is applied to assign weights to each feature, and threshold-based analysis is used to detect fall events. Simulation results on the Le2i Fall Detection Dataset demonstrate that the proposed method achieves an accuracy of 88%, confirming its effectiveness in real-world applications.