Skip to main content
. 2025 Aug 12;16:1649330. doi: 10.3389/fphys.2025.1649330

TABLE 1.

Provides a comparative overview of depth cameras and pose estimation frameworks, reviewing their typical error margins, envir onmental constraints, reliability, and suitability for recording players’ performance, with a focus on body limbs, setup complexity, and suitability for recording.

System Type Error range Environment suitability Depth range (m) 2D/3d output Reliability (limb Focus) Setup complexity Key strengths and limitations
Kinect v2 Depth Camera ∼10°–20° (Pfister et al., 2014) Indoor only 0.5–4.5 (Kurillo et al., 2022) 3D Both (higher in the sagittal plane) Low Built-in skeletal tracking
ZED 2i Stereo Camera ∼5°–15° (Aharony et al., 2024) Indoor/Outdoor, >3 m 0.2–20 (Aharony et al., 2024) 3D Lower limb (wide FOV applications) Medium-High Higher performance for wide views
RealSense Depth Camera ∼10°–20° (Pilla-Barroso et al., 2024) Indoor/Outdoor 0.2–10 (Aharony et al., 2024) 2D/3D Lower limb (slow gait, squat) Medium Valid for close-range tracking
OpenPose Software ∼5°–15° (Ino et al., 2024) Indoor/Outdoor Camera-limited 2D (pseudo-3D) Upper limb (fine movement detection) Medium Good multi-person support
MediaPipe Software ∼<10° (Pilla-Barroso et al., 2024) Indoor/Outdoor Camera-limited 2D Lower limb (squats, gait) Low Fastest runtime, low memory use, suitable for mobile apps
AlphaPose Software ∼15°–25° (Hulleck et al., 2023) Indoor/Outdoor Camera-limited 2D Upper limb (combat/sport gestures) Medium Higher RMSE in some studies
DensePose Software ∼10°–20° (Suzuki et al., 2024) Indoor preferred Camera-limited 2D/3D mesh Both (AR/VR mesh mapping) High Ideal for multi-person and VR/AR

The numbers refer to the references mentioned in the text.