A Sequential Approach to 3D Human Pose Estimation: Separation of Localization and Identification of Body Joints
In this paper, we propose a new approach to 3D human pose estimation from a single depth image. Conventionally, 3D human pose estimation is formulated as a detection problem of the desired list of body joints. Most of the previous methods attempted to sim
- PDF / 2,138,276 Bytes
- 15 Pages / 439.37 x 666.142 pts Page_size
- 42 Downloads / 194 Views
2
Department of ECE, ASRI, Seoul National University, Seoul, Korea {n12345,mks0601,kyoungmu}@snu.ac.kr Division of CESE, Hankuk University of Foreign Studies, Yongin-si, Korea [email protected]
Abstract. In this paper, we propose a new approach to 3D human pose estimation from a single depth image. Conventionally, 3D human pose estimation is formulated as a detection problem of the desired list of body joints. Most of the previous methods attempted to simultaneously localize and identify body joints, with the expectation that the accomplishment of one task would facilitate the accomplishment of the other. However, we believe that identification hampers localization; therefore, the two tasks should be solved separately for enhanced pose estimation performance. We propose a two-stage framework that initially estimates all the locations of joints and subsequently identifies the estimated joints for a specific pose. The locations of joints are estimated by regressing K closest joints from every pixel with the use of a random tree. The identification of joints are realized by transferring labels from a retrieved nearest exemplar model. Once the 3D configuration of all the joints is derived, identification becomes much easier than when it is done simultaneously with localization, exploiting the reduced solution space. Our proposed method achieves significant performance gain on pose estimation accuracy, thereby improving both localization and identification. Experimental results show that the proposed method exhibits an accuracy significantly higher than those of previous approaches that simultaneously localize and identify the body parts. Keywords: Depth camera
1
· Human pose · Regression forest
Introduction
Real-time 3D human pose estimation is a core technique for activity recognition. It has diverse applications including human-computer-interface development, teleoperation, health monitoring, and learning by demonstration(LbD) [1]. Conventionally, 3D human pose estimation is formulated as a detection problem of the desired list of body joints. Most of the previous methods simultaneously localize and identify joints, such as localization of left shoulder, with the expectation that the accomplishment of one task would facilitate the accomplishment c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part V, LNCS 9909, pp. 747–761, 2016. DOI: 10.1007/978-3-319-46454-1 45
748
H.Y. Jung et al.
Fig. 1. Similar local depth features are exhibited by different body parts at various poses. Thus, a local depth feature may be insufficiently distinctive to distinguish the identity of parts
of other [2–5]. By contrast, we assert in this paper that identification hampers localization; therefore, the two tasks are treated separately and significant performance boost is obtained. In depth images, similar local patch patterns are easily observed from different body parts at various poses. For example, in Fig. 1, right arm of the first person, left arm of the second person, and left leg of the third person i
Data Loading...
