Research On Skeleton Localization Based On Kinect Sensor

With the evolvement of the techniques in the fields of pattern recognition and artificialintelligence, the human-computer interaction technology has been employed in a wide variety ofareas. In recent years, body motion analysis has progressed prominently, which is dependent onthe prior task where body motion is captured. Skeleton model extraction as a core chain in thisprocess has received extensive focus from the academia. Skeleton is a significant feature thatretains the image topology, and is an effective form for shape representation of an object. Beingable to represent the geometric properties of an object with a small amount of information, it iswidely adopted in the areas of pattern recognition, image retrieval, virtual movement, etc. Thispaper carries out the following work based on the research of Kinect skeleton positioning:First of all, the recent research state of skeleton localization is summarized, its merits anddrawbacks are analysed, the background of this paper is illustrated.Secondly, the hardware system, software system, the working principles of the Kinectsensor and decision tree based machine learning algorithm are introduced.Again, a segmentation method based on optimal threshold is put forward, which is utilisedto segment body from the background; mathematical morphology is employed to denoise thesilhouette of the body and extract the skeleton by incorporating the Hilditch thinning algorithm.Lastly, this paper resolves the obstruction issue caused by the arms by taking advantage ofthe features of the depth map. The complete body movements are captured by maximisingtriangular area. The results of the experiments validated the effectiveness of this scheme.