Research on spatial motion distance features of human body joint points based on dual-video acquisition experimental environment

To address the technical challenges in individual identification via video surveillance for criminal investigation, this study proposes a feature recognition method based on spatial distances of human joint points. Utilizing the Simi Motion 3D motion capture system, we collected 400 gait cycle datasets from 20 homogeneous participants and established a feature model incorporating 18 joint relative distances. Through coefficient of variation analysis and principal component evaluation, eight high-distinctiveness feature points were selected. An innovative Feature Recognition Reliability Index (FRRI) system was designed, combined with a three-level adaptive weight fusion strategy, which significantly improved recognition accuracy in complex environments. Experimental protocol: 10-fold cross-validation was adopted, dividing 400 sets of data into a training set (320 sets) and a test set (80 sets) at an 8:2 ratio. Feature selection (PCA/FRRI) was performed within training folds to avoid optimistic bias. Experimental results verified the reliability of different gait features. This research provides a quantifiable forensic technical solution for gait-based individual identification of criminal suspects, demonstrating practical value in judicial authentication and intelligent security applications.