Research On3D Reconstruction Based On Database

Database-based human motion capture is one of the important topics of theresearch field of computer vision, but also in recent years been a concern toinvestigators forefront of direction. The paper focuses on the database-based motioncapture, including search and query of action, action database storage architecture aswell as the three-dimensional reconstruction of the human body.The rapid growth of the action of the motion capture data reuse becomesincreasingly important. Based on the database, the user can input data query system toobtain the action that they want, which improve efficiency.This article provides two ways to get the input data of the system. First, thesystem provides a user interface. Second, may be a monocular video as the input.Then find before we processed the data, making retrieval fast. Therefore, we use thePCA extract action feature vectors, and then establish a one-to-one relationship ofAction and eigenvectors. The feature vector of the input data also extract the sameoperation, so that to match the action database, to reduce the time complexity. Takinginto account that we want to achieve real-time effects during crawling parallelcomputing. Use the GPU for hardware acceleration, so that when the database is large,computing time is low.Finally, based on a threshold value of the sequence of a result if input singleframe, then we choose the closest one. If the input is a sequence of frames then thisarticle should choose a frame sequence with frame sequence matching algorithm toget a match, so that the resulting sequence as smooth as possible, as close as possibleto the original sequence. Database storage has not been taken seriously, when a lot ofaction, we have to consider using the optimized architecture to store, reducing thespace complexity. With respect to store each frame of action of the joint coordinatesof the point, we can go to the store eigenvectors, and then use the idea of the SVDmatrix compression, so good control of the size of the database.Experimental results show that the proposed algorithm can better solve themotion capture in the presence of the key issues for further research laid thetheoretical foundation.