One of the most interesting things we can do with our GaMoCap is computing joint angles for human motion; that means compute a real skeleton model for humans. Since we had the opportunity to use a Vicon system for few days, we decided to acquire some sequences and compare results. The following is what we did.
Two systems, an eight cameras Vicon system and our Garment MoCap have been used to acquire some sequences of humans moving in a small room. In particular we focused on some standard simple movements involving just two or three limbs, in order to have just one or two articulations working at the same time. One of this movements, we will use as example, is the hip adduction: starting from a standing position the subject raises the extended leg in a lateral path as much as possible, in this kind of motion the hip is working as a single axis rotational joint.
Figure 1: Hip adduction schematic motion
There are two possible layouts for Vicon markers:
- Three markers layout: two markers on the limbs (torso, thigh) + one on the joint (hip);
- Twelve markers layout: three for each limb (torso, thigh, shin) + one at each joint;
The second is in general more robust than the first, since the three markers layout requires a very accurate positioning of markers, while in the second layout the markers can be random positioned.
In this test we considered just two limbs: torso (with 3 markers) and all the rest (8 markers). The marker on the hip is not taken into account since its motion is a composition of motions of two limbs.
Figure 2: Subject wearing garment and Vicon markers disposition (A), and Vicon skeleton of markers (B)
3D points cloud recovered by GaMoCap is then processed to segment different limbs of human body; this is performed automatically by means of Local Subspace Affinity (LSA) algorithm. In this sequence we segmented three bodies (torso and both legs) but we are just interested in angle between bodies: torso and right leg.
Figure 3: Segmentation with LSA: left leg (body 1, red), torso (body 2, green) and right leg (body 3, blue)
In both cases, with Vicon and GaMoCap, we can use factorization algorithms to compute the motion of each limb considered as a rigid body. Composing this motions we can also compute the relative motion between two bodies and straightforward the joint angle, approximating the hip as a single axis rotational joint. Comparison results are provided in terms of absolute angle and residuals.
Figure 4. Angle estimation and absolute difference between the estimated angles.