Skeleton Model Based Actor Segmentation in Point Clouds Generated by 3D Range Sensors

OData support
Supervisor:
Dr. Szécsi László
Department of Control Engineering and Information Technology

The exact observation of human movement is an important in respect of health, sport, movie making and entertainment industries. Describing the movement of certain limb segments is a good model for this. The tracking of the segments can be achieved with camera systems, ultrasound transmitters and movement sensors, but the attached devices and the special clothing can distort the measurements. Therefore nowadays markerless skeleton tracking is an actively researched scientific discipline.

In addition to the 2D image sensors, 3D (rather: 2,5D) sensors are more and more available to common usage, due to the increasingly cost-effective and precise manufacturing technologies. These sensors grant us the ability to observe and analyse the space surrounding us (and the interactions happening within it) in increasingly more detail. Separating the observed objects based on the depth information obtained by these sensors is much easier than using only the usual colour and texture data. The results of this segmentation were applied to solve a well-defined practical problem.

The results of the segmentation were used in a practical task.

Sport is one of the most important ways to preserve one’s health. Trying out a sport at home or simply exercising to improve skills helps people enjoy sports as well. (Naturally, overview by a skilled coach is still necessary to reach optimal results.)

In my dissertation I examined the skeleton-based comparison to improve trainings at home. With my application will be able to compare the trainer’s and the sportsman’s movements for improving the home workout.

Downloads

Please sign in to download the files of this thesis.