Font Size: a A A

The Research And Implementation On Human Motion Tracking Based On Global Optimization

Posted on:2014-12-25Degree:MasterType:Thesis
Country:ChinaCandidate:B LiFull Text:PDF
GTID:2268330425483902Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
With the rapid development of areas such as the analysis of human-computerinteraction、3D animation、games、sports、medical diagnostics and virtual reality,human motion capture system is a key technology and becomes the hot areas ofresearch. On the market, human motions capture system using the marks has beenvery mature and commercialization. Researchers have begun focusing on marker-lesshuman motion capture system. Currently, research on it is still at an early stage andthe technology is in itsinfancy. Cameras calibration and key point tracking are mainlytask in this paper about human motion capture system using marker-less.The main innovative research work of this paper can be summarized as follows:According to independence of objects, a parallel algorithm about thefundamental matrix based on random sample consensus is presented. It is a key step incamera calibration process. Random sample consensus (RANSAC) is usually used inthe fundamental matrix algorithm. The the traditional random sample sonsensus is therepeated process of the “assumptions–verification”. We introduce an improvedmethod with assumptions and validation process. A parallel algorithm is proposed forthe calculation of the fundamental matrix.In light of sparse matrix in the optimization process, we get low-dimensionalincremental equation from high-dimensional space. Bundle adjustment is the basicmethod for multi-camera parameter optimization. The key step in bundle adjustmentis how to update the independent variable. The traditional algorithm updated theindependent variable by soloving a high-dimensional matrix. According to thesparsity of the coefficient matrix, an algorithm using low-dimensional equationsinstead of high-dimensional equation is proposed and implemented.In human motion capture system, a key point tracking algorithm based on globaloptimization is introduced. The traditional marker-less human motion capture used atracking algorithm based on image segmentation. This method exist some shortage,such as action inaccurate and inconsistent data. It is also easy to slip and drift. Keypoints trajectory based on the global optimization algorithm is more smooth andcontinuous. Reconstruction of human body joint motion is relative stable andcontinuous. Even key points is occlusion situation, the algorithm based on globaloptimization is still working. The virtual character’s movement is driven by3D reconstruction. Experimental results show that the virtual character’s movement iscoherence and natural.
Keywords/Search Tags:random sample consensus (RANSAC), nonlinear optimization, cameras calibration, key point tracking, human motion capture
PDF Full Text Request
Related items