Table 3.
Markerless motion capture devices and feature extraction methods used.
| Study | Device (camera or sensor) | Devices, n | Frames per second (fps or Hz) | Position of cameras | Extraction methods |
| Cimolin et al (2022) [24] | Kinect V2 | 1 | 30 | Tripod in front | Custom algorithm |
| Kaur et al (2023) [25] | Video camera | 2 | 30 | Front and right side | OpenPose |
| Khan et al (2021) [26] | Video camera | 1 | 25 | Front | Custom algorithm |
| Khan et al (2013) [27] | Video camera | 1 | 5 | Not reported | Custom algorithm |
| Kondragunta et al (2020) [19] | Kinect V2 | 1 | 20 | Not reported | OpenPose |
| Lai et al (2022) [28] | Kinect V2 | Not reported | 30 | Not reported | GaitBEST (LongGood Meditech) |
| Li et al (2018) [29] | Video camera | 1 | 25 | Not reported | Iterative Error Feedback and OpenPose |
| Mehdizadeh et al (2021) [9] | Kinect V2 | 1 | Not reported | Ceiling in hallway | Custom algorithm |
| Mehdizadeh et al (2021) [15] | Kinect V2 | 1 | Not reported | Ceiling in hallway | Custom algorithm |
| Morinan et al (2022) [30] | Mobile phone camera and KELVIN-PD (Machine Medicine) mobile app | Not reported | Not reported | Not reported | OpenPose |
| Munoz-Ospina et al (2022) [31] | Kinect eMotion | 1 | Not reported | Not reported | Custom algorithm |
| Ng et al (2020) [16] | Kinect V2 | 1 | 30 | Ceiling at the end of a hallway | OpenPose |
| Ospina et al (2021) [32] | Kinect V1 | 1 | Not reported | Participants walking toward the camera | Custom algorithm |
| Otte et al (2020) [33] | Kinect V1 | 1 | 30 | 1.4 m height in front | Custom algorithm |
| Pedro et al (2020) [34] | Azure Kinect | 3 | 30 | Each end of walkway and halfway between | Azure Kinect SDK to extract joint positions to estimate 32 body joint poses from depth color recordings |
| Procházka et al (2015) [35] | Kinect V1 | 1 | 30 | 60 cm above floor | Custom algorithm |
| Rupprechter et al (2021) [36] | Mobile phone camera and KELVIN-PD (Machine Medicine) mobile app | 1 | Not reported | Patients walking directly toward or away from the camera in hallways or office settings | OpenPose |
| Sabo et al (2022) [37] | Logitech C920 | 1 | 30 | Tripod mounted, at one end of walkway | AlphaPose (Shanghai Jiao Tong University), Detectron (Facebook AI Research), OpenPose, and ROMP (Regress All Meshes in a One-Stage Fashion for Multiple 3D People; JD AI research) |
| Sabo et al (2022) [18] | Kinect V2 and mobile phone cameras | 1 | 30 | Kinect: hallway ceiling; stationary mobile phone camera: participants walked toward and away from | OpenPose, Detectron, and AlphaPose |
| Sabo et al (2021) [17] | Kinect V2 | 1 | 30 | Ceiling in hallway | AlphaPose and engineered 2D gait features from joint trajectories |
| Sabo et al (2020) [38] | Kinect V2 | 1 | 30 | Ceiling in hallway | OpenPose |
| Seifallahi et al (2022) [39] | Kinect V2 | 1 | Not reported | On a tripod at a suitable distance from an oval path | Custom algorithm |
| Shin et al (2021) [40] | Video camera | 1 | 30 | Frontal view from a tripod-mounted camera 1.5 m from the horizontal line of the turning point | OpenPose, OpenCV |
| Soltaninejad et al (2018) [41] | Kinect V2 | Not reported | 30 | Not reported | Graph model of body skeleton |
| Tan et al (2019) [42] | Kinect V2 | 1 | Not reported | End of walkway | Custom algorithm |
| Ťupa et al (2015) [43] | Kinect V1 | 1 | 30 | 60 cm above floor | Custom algorithm |