Commercial Applications of 3D Human Pose Estimation in Biotechnology: A Two-Stage Fusion with Multi-Feature Integration Approach

Abstract

Three-dimensional human pose estimation (3D-HPE) has rapidly advanced as a critical area of research within computer vision, demonstrating significant potential for biotechnological applications. This paper introduces an innovative approach to enhance the accuracy of 3D-HPE, utilizing a two-stage model with a multi-feature fusion technique. The model employs convolutional kernels of varying sizes to extract feature maps from different receptive fields, resolutions, and dimensions. Initially, these feature maps are fused with the 2D coordinates of key joint points from the input frame. Subsequently, in the second stage, the fused feature maps are integrated with the feature points of 2D key joints to accurately predict the 3D key joints of the human body. Experimental analyses on the Human3.6M dataset indicate that our proposed method significantly outperforms existing approaches, achieving improvements of 9.47% in Mean Per Joint Position Error (MPJPE) and 8.55% in Procrustes-aligned MPJPE. These metrics are essential for gauging the precision of pose estimations, crucial in applications such as human-computer interaction (HCI), motion analysis, and virtual reality—areas of growing interest within the biotechnology sector. The two-stage model with multi-feature fusion not only offers a more precise method for 3D-HPE but also opens avenues for commercial applications, particularly in medical diagnostics, physical therapy, and personalized healthcare solutions. The results underscore the commercial viability and significant potential of the proposed approach to contribute effectively to advancements in biotechnological practices, enhancing both the accuracy and applicability of 3D pose estimation in real-world scenarios.