Quantification of Occlusion Handling Capability of 3D Human Pose Estimation Framework

3D human pose estimation using monocular images is an important yet challenging task. Existing detection methods exhibit excellent performance under normal conditions however their may degrade due to occlusion. Recently some occlusion aware have also been proposed however, the handling capability of these networks has not thoroughly investigated. In current work, we propose occlusion-guided framework and quantify its by different protocols. The method estimates more accurate poses 2D skeletons with missing joints as input. Missing are handled introducing guidance that provides extra information about absence or presence a joint. Temporal exploited better estimate joints. A large number experiments performed for quantification on three publicly available datasets in various settings including random joints, fixed body parts missing, complete frames mean per joint position error criterion. addition that, quality predicted evaluated action classification estimated achieved significantly improved recognition Our demonstrate effectiveness well deep neural networks.