Personalized Virtual Human 3D Modeling Based On Multiple Kinect

3D modeling of personalized virtual human is a technology of creating a user’s 3D personalized geometric modeling a virtual scene. This article uses multiple Microsoft Kinect depth cameras to complete the reconstruction of user’s model. The workflow is as follows: First, according to the characteristics of Kinect camera we choose the appropriate camera number and location to build a low-cost data acquisition system to capture the virtual humans; secondly, using a piece of unfolded thickness paper with many creases, we calibrate the Kinect cameras with iterative closest point algorithm, and prepare for the global registration of the local data; again, multiple cameras simultaneously and automatically capture static human subjects to be built from multiple perspectives, each camera automatically obtain the depth point clouds of human subjects, taking into account the limitations of Kinect camera and the point cloud data’s integrity and high noise, we use bilateral filtering method for data pre-processing to compensate for defects and smooth noise reduction purpose,and re-use camera calibration step information on completion of the initial global point cloud registration, then we use the global data feature matching method for global optimization to get an exact match point cloud; finally, using the Poisson surface reconstruction method to reconstruct the model’s surface, we simplify the Poisson’s equation into a sparse linear system in solving the Poisson’s equation, and then use the MultiGrid method for iterative conjugate linear equation solving to achieve personalized virtual human 3D modeling requirements.Experimental results show that this article can reconstruct the high-fidelity personalized 3D human model. Compared with the same type of work, it’s very easy to use, no turntable and other equipment, lower cost and easier to build the system; we can achieve real-time data collection, the data acquisition time lasts short than 1 second, users do not need long time to remain stationary state anymore; reconstruction time is about 90 seconds, shorter than other reconstruction time, reconstruction effect has high accuracy, and response the user’s individual characteristics better.