Validation of lower limb joint center localization and kinematics in markerless motion capture during sloped walking

For decades, scientists studying how humans move have relied on a method that looks like a high-tech version of a game of connect-the-dots. To understand the mechanics of a walking step, researchers attach small reflective markers to a person's skin and use cameras to track their positions. This technique, known as optical motion capture, has long been the gold standard for measuring joint angles and movement patterns. However, it has a significant flaw: the skin moves independently of the bones underneath. As muscles contract and shift during a step, the markers on the surface slide, creating a blur of data that does not perfectly match the true motion of the skeleton. This problem, called soft tissue artifact, becomes especially difficult to solve when a person is walking on steep hills or performing complex movements, as the skin stretches and bulges more dramatically.

In recent years, a new approach has emerged that skips the markers entirely. Using advanced computer vision and deep learning, these markerless systems analyze video footage to estimate where the joints are located based on the shape of the body itself. While this technology promises to make movement analysis more natural and accessible, scientists have been cautious about how well it holds up under pressure. The question remains whether these digital estimates can match the precision of the traditional marker-based systems, particularly when the body is under the strain of walking on steep slopes.

A team of researchers in Japan set out to test the limits of this new technology by putting it through a rigorous trial on a treadmill. They recruited twenty-four healthy adults and asked them to walk at a steady pace across five different conditions: flat ground, moderate uphill and downhill slopes, and steep uphill and downhill slopes. To ensure a fair comparison, the participants wore tight-fitting suits that minimized the loose fabric that could confuse the cameras, and they walked while being recorded simultaneously by two different systems. One system used the traditional reflective markers and infrared cameras, while the other used eight standard video cameras to capture the movement without any physical attachments. The researchers then used powerful software to calculate the exact position of the hip, knee, and ankle joints, as well as the angles of the legs, for every moment of the walking cycle.

The results revealed a story of both remarkable success and specific, predictable limitations. When the researchers looked at the actual location of the joint centers in three-dimensional space, the markerless system performed with extraordinary precision. The estimated positions of the hips, knees, and ankles matched the traditional system almost perfectly, regardless of whether the person was walking on flat ground or a steep fifteen-percent slope. This suggests that the computer algorithms are exceptionally good at finding the center of a joint, even when the body is moving quickly or the view is partially blocked by other body parts.

However, the story changed when the researchers examined the angles of the joints. While the overall shape of the movement curves was very similar between the two systems, there were consistent differences in how the angles were reported. For the hip and knee moving forward and backward, the two systems agreed closely, but the markerless system consistently reported a slightly different starting tilt for the pelvis. This small, constant offset meant that while the pattern of movement was correct, the absolute numbers were shifted. The differences became more pronounced in the side-to-side and twisting motions of the joints. The traditional system, which relies on markers stuck to the skin, showed more erratic fluctuations, likely because the skin was sliding over the bone. In contrast, the markerless system produced smoother data, but this smoothness sometimes masked the true, complex movements of the bones, leading to significant disagreements in how much the hip and ankle were rotating or tilting sideways.

The most striking difference appeared when the participants walked downhill. In this condition, the muscles must work harder to control the body's descent and absorb the shock of landing. The researchers found that the disagreement between the two systems grew larger during downhill walking, particularly for the hip's side-to-side motion. They suspect that the intense muscle activity and the impact of landing caused the skin on the thigh to move and vibrate in ways that confused the traditional marker system, while the markerless system, trained on average movement patterns, smoothed over these extreme variations. This suggests that while the new technology is robust enough to track joint locations on steep terrain, the specific angles it calculates for twisting and side-to-side movements may need careful interpretation, especially when the body is under high physical stress.

Ultimately, the study confirms that markerless motion capture is a powerful tool for analyzing movement in challenging environments where traditional methods might struggle with equipment or occlusion. The technology can reliably track where the joints are, making it suitable for field studies and complex tasks. Yet, the findings serve as a reminder that the numbers it produces for joint angles are not identical to those from the older, marker-based methods. Scientists and clinicians using this new technology will need to account for these systematic differences, particularly when analyzing the twisting and tilting of the body during demanding activities like walking down a steep hill.