Research On3D Human Model Registration Technology Based On Kinect

A critical step in the process of treating the burn patients is to accurately evaluate its degree of injury. Timely and precise calculation of the total burned area is the foundation to provide the appropriate amount of fluid. It is also the basis to determine the further therapeutic scheme. So, it is necessary to research an efficient and fast the burned area measurement method. Based on this, this research project proposed to achieve a digital automatic system for evaluating the burns of human body. The implementation process of this system mainly includes the following four steps: data acquisition,3D model registration, burned region extraction, and total area statistics.This article mainly studies the part of three-dimensional model registration. On the issue that the traditional3D body scanning method and registration techniques cannot be applied to the aforesaid system very well, this article proposed a new3D registration method based on the front and back segments of human body for grey model reconstruction, and another registration method combined with the color character for color model reconstruction. Experiments show that these two registration algorithms are suitable for this system. They improved the efficiency and accuracy of registration compared with the traditional algorithm.On the study of color model reconstruction, we learned that it is uneasy to obtain the model data of burned patients under various angles because of its inconvenience of moving. So, completed3D human body model cannot be achieved by the traditional3D registration methods. This paper proposed a registration method to create a whole3D human body model by just using its front and back segments without any overlapping area, based on the improvement of scanning way. This method uses the contours of these two segments as character points, and receives a great result after registration and gap merge process.For the reason that Kinect can obtain the scanning object’s spatial position data as well as its color character, this paper put forward the registration algorithm by taking full use of the obtained color information. This method added a comparison process of each feature point’s color value during searching the corresponding characteristic points. It improves the accuracy of the corresponding feature points. At the same time, color information in RGB space is converted into HSV space for the effect of light intensity. Experiments show that this method can not only reduce the iterations, but also get a higher precision result compared with the traditional Iterative Closest Point (ICP) algorithm.Finally, this paper develops a non-real-time continuous3D model registration system, and gives the related experimental results and analysis.