Technique developed to capture human movement in 3D

Engineer Pedro Correa, from the UCL Telecommunications and Teledetection Laboratory, told SINC that, together with professor Ferran Marqués's unit at the UPC, they have developed algorithms that tackle the problem of gesture recognition “in the least invasive way possible, since it does not require wearing any special suit or receivers, using a simple video camera to film the body's movement”.

The images filmed identify the person's outline several dozens of times a second, and the data obtained are analyzed by the algorithm invented by the researchers to identify the “crucial points”: head, hands and feet. The “crucial points extraction algorithm” uses the mathematical concept of geodesic distance to calculate the person's extremities, “in other words”, clarifies Correa, “which points are furthest away from the center of gravity, following a path entirely within the outline”.

Once the extremities have been obtained, the outline is analyzed once again to create “morphological skeletons” that help assign a label to each extremity. The five possible labels are head, left hand, right hand, left foot and right foot. Once identified, they are represented with colored dots for tracking in 2 dimensions. This enables the user to analyze the results visually.

To obtain the same information in 3 dimensions, the same steps are taken with an additional camera. This way, the triangulation of the labels extracted in each of the two views makes it possible to obtain the points in a three dimensional space. The front view provides information on the vertical and horizontal positions of the extremities, and the side view provides information on their depth.

The low level of complexity in this system allows it to be applied in real time on any personal computer, with a margin of error of between 4% and 9% in real situations, depending on the context and the quality of the segmentation carried out.

Correa explained that the applications of this technique are “all those that require motion interaction with the computer; that is, from browsing through applications in an operating system (like moving windows and text with hand movements) to interactive aerobic video games, and much more”. The study was also participated in by a Belgian company specializing in real-size video games, which are used, for example, in amusement parks and museums.