Maximizing Markerless Motion Capture Data Processing Speed: Enabling Efficient Reporting

Markerless motion capture technologies are rapidly being developed and applied in research. Even with the advantages of markerless motion capture over marker-based motion capture (i.e., faster participant set-up, higher portability), data analysis time can restrict researchers’ ability to check data quality in real-time and limits real-time feedback for participants. PURPOSE: To quantify the effect of data collection and data processing methods and equipment on markerless motion capture data analysis time. METHODS: Markerless motion capture data were collected with four participants completing three trials of walking, overhead squats (OHS), and countermovement jumps (CMJ) for 30 seconds and recorded at 120 Hz using an 8-camera system. To evaluate the effect of data collection and processing methods and equipment on analysis time, the same trials were analyzed multiple ways: 1, 2, 3, or 4 participants identified; 10, 20, or 30 second trial length; walking, OHS, or CMJ movements; at 60 Hz or 120 Hz; and with three different graphics processing unit (GPU) configurations. Multivariable regression was used to evaluate how each variable affected data analysis time (α-level = 0.05). RESULTS: Data analysis time significantly increased as trial length (p < 0.001) and sample rate (p < 0.001) increased, and significantly decreased as GPU quality and quantity increased (p < 0.001). Data analysis time was not affected by movement (p = 0.112) or by the number of participants identified (p = 0.440). Data are displayed in Figure 1. CONCLUSION: Markerless motion capture analysis time can be maximized by recording multiple participants simultaneously. Researchers should collect data with the shortest trial lengths and lowest possible sample rates while utilizing as many high-quality GPUs as possible. In the fastest configuration, a 10 second walking trial required 65 seconds to analyze. This is nearly a feasible solution for real-time data quality control and participant feedback. Supported by: This research was partially funded by an NSF Industry/University Cooperative Research Program called the Center for Disruptive Musculoskeletal Innovations (IIP-1916629).