Research And Application Of Hand Center Positioning Method Based On Kinect

Nowadays Kinect has been widely used in education, health care, Fast MovingConsumer Goods, daily life and many other fields. But the stability and accuracy ofclose identification cannot meet the needs of human-computer interaction. In orderto enhance the precision and practicality of gesture interactive application, the studyof Kinect hand positioning has become necessary.A series of pretreatment for hand positioning is detailed in this dissertation. Atransformation formula is put forward to calibrate the output of depth information. Agray dynamic compensation module written by myself is added in human extractionto cover the mismatch of depth and RGB, then fixed threshold is used to strengthenthe hand segmentation flexibility.There are two kinds of hand features: one is microscopic like hand contours,convex hull, fingertips and so on, another is hand-center positioning on behalf ofmacro feature. Hand micro feature are extracted by computer vision algorithms inthis dissertation, the realization of Euclidean Distance Transform hand centerpositioning is finished, then its advantage and SDK’s weakness is evaluated.There is no mathematical formula to define hand center point at present. A wayis proposed in this dissertation to get the real position of hand center which meet theneed of human-computer interaction. With the help of self-created evaluation index–improvement rate, Euclidean Distance Transform is proved to be advanced instability and accuracy over SDK in quantitative discussion. It is significantespecially for close range, different hand shape and angles, both in distance anddirection.At the same time, Euclidean Distance Transform is applied to the whole bodyimage. It can localize shoulder center joint and it’s also stable. Finally an Anglemeter application is developed. It has Education and entertainment meaning, verifythe practicability of the Euclidean distance transform.