In surveillance tasks we would like to get more details about the persons tracked by deployed cameras. For this, the conventional cameras not always give satisfactory results, so the use of thermal and range cameras is spreading. These cameras can operate successfully in disadvantaging conditions.
In my work I used two different kinds of range cameras, which are capable of producing three-dimensional point cloud sequences of the watched area. Both devices are active cameras, which scan the surrounding area by emitting infrared light. The first sensor is the SwissRanger 4000 manufactured by Mesa Imaging AG. This sensor is a Time-Of-Flight camera, which measures the distance of an object by measuring the flight time of light. The second device is the Kinect sensor of Microsoft, which measures spatial depth with structured light, with a relatively high resolution. Based on the sensors point cloud data, the segmentation of actors can be done, and skeletal models can be applied onto them. The model estimates the important joints of the actor, which describes the person’s pose and position. This position data cannot be used in its original form, therefore deriving of static and dynamic parameters are required which describe the state of skeleton more precisely. Based on the derived parameters we can create binary features from the actual state of skeleton. Then based on these features we can decide the actor in which actions and interactions are participating. I have integrated the functions above into an application with graphical user interface, which helps to monitor the process, and observe skeleton parameters.