Consistent Correspondences for Shape and Image Problems

Establish consistent correspondences between different objects is a classic problem in computer science/vision. It helps to match highly similar objects in both 3D and 2D domain. Inthe 3D domain, finding consistent correspondences has been studying for more than 20 yearsand it is still a hot topic. In 2D domain, consistent correspondences can also help in puzzlesolving. However, only a few works are focused on this approach. In this thesis, we focuson finding consistent correspondences and extend to develop robust matching techniques inboth 3D shape segments and 2D puzzle solving. In the 3D domain, segment-wise matching isan important research problem that supports higher-level understanding of shapes in geometryprocessing. Many existing segment-wise matching techniques assume perfect input segmentation and would suffer from imperfect or over-segmented input. To handle this shortcoming,we propose multi-layer graphs (MLGs) to represent possible arrangements of partially mergedsegments of input shapes. We then adapt the diffusion pruning technique on the MLGs to findconsistent segment-wise matching. To obtain high-quality matching, we develop our own voting step which is able to remove inconsistent results, for finding hierarchically consistent correspondences as final output. We evaluate our technique with both quantitative and qualitativeexperiments on both man-made and deformable shapes. Experimental results demonstrate theeffectiveness of our technique when compared to two state-of-art methods. In the 2D domain,solving jigsaw puzzles is also a classic problem in computer vision with various applications.Over the past decades, many useful approaches have been introduced. Most existing worksuse edge-wise similarity measures for assembling puzzles with square pieces of the same size, and recent work innovates to use the loop constraint to improve efficiency and accuracy. Weobserve that most existing techniques cannot be easily extended to puzzles with rectangularpieces of arbitrary sizes, and no existing loop constraints can be used to model such challenging scenarios. We propose new matching approaches based on sub-edges/corners, modelledusing the MatchLift or diffusion framework to solve square puzzles with cycle consistency.We demonstrate the robustness of our approaches by comparing our methods with state-of-artmethods. We also show how puzzles with rectangular pieces of arbitrary sizes, or puzzles withtriangular and square pieces can be solved by our techniques.

Hierarchical Clustering-Aligning Framework Based Fast Large-Scale 3D Reconstruction Using Aerial Imagery

With extensive applications of Unmanned Aircraft Vehicle (UAV) in the field of remotesensing, 3D reconstruction using aerial images has been a vibrant area of research. However,fast large-scale 3D reconstruction is a challenging task. For aerial image datasets, large scale meansthat the number and resolution of images are enormous, which brings significant computationalcost to the 3D reconstruction, especially in the process of Structure from Motion (SfM). In thispaper, for fast large-scale SfM, we propose a clustering-aligning framework that hierarchicallymerges partial structures to reconstruct the full scene. Through image clustering, an overlappingrelationship between image subsets is established. With the overlapping relationship, we proposea similarity transformation estimation method based on joint camera poses of common images.Finally, we introduce the closed-loop constraint and propose a similarity transformation-based hybridoptimization method to make the merged complete scene seamless. The advantage of the proposedmethod is a significant efficiency improvement without a marginal loss in accuracy. Experimentalresults on the Qinling dataset captured over Qinling mountain covering 57 square kilometersdemonstrate the efficiency and robustness of the proposed method.

General Formula of Mobility Calculation for Planar Mechanism

Deficiencies are existed for currently formulas of mobility calculation for planar mechanism. They are not suitable for planar mechanism with virtual constraints and the number of general constraints equal to 4. To solve the problem, the new concepts of virtual loop, virtual-loop constraint and virtual pair are defined to establish a general f ormula for DOF of planar mechanism; the calculation method for virtual-loop constraint and the mobility of link-group are also given. It is proved that the new formula is correct, general, simple and effective through the mobility analysis of several different kinds of planar mechanisms.

Vibration Calculation of Spatial Multibody Systems Based on Constraint-Topology Transformation

ABSTRACTMany kinds of mechanical systems can be modeled as spatial rigid multibody systems (SR-MBS), which consist of a set of rigid bodies interconnected by joints, springs and dampers. Vibration calculation of SR-MBS is conventionally conducted by approximately linearizing the nonlinear equations of motion and constraint, which is very complicated and inconvenient for sensitivity analysis. A new algorithm based on constraint-topology transformation is presented to derive the oscillatory differential equations in three steps, that is, vibration equations for free SR-MBS are derived using Lagrangian method at first; then, an open-loop constraint matrix is derived to obtain the vibration equations for open-loop SR-MBS via quadric transformation; finally, a cut-joint constraint matrix is derived to obtain the vibration equations for closed-loop SR-MBS via quadric transformation. Through mentioned above, the vibration calculation can be significantly simplified and the sensitivity analysis can be conducted conveniently. The correctness of the proposed method has been verified by numerical experiments in comparison with the traditional approaches.

Integrated Kinematic Calibration for All the Parameters of a Planar 2DOF Redundantly Actuated Parallel Manipulator

In this paper, the kinematic calibration of a planar two-degree-of-freedom redundantly actuated parallel manipulator is studied without any assumption on parameters. A cost function based on closed-loop constraint equations is first formulated. Using plane geometry theory, we analyze the pose transformations that bring infinite solutions and present a kinematic calibration integrated of closed-loop and open-loop methods. In the integrated method, the closed-loop calibration solves all the solutions that fit the constraint equations, and the open-loop calibration guarantees the uniqueness of the solution. In the experiments, differential evolution is applied to compute the solution set, for its advantages in computing multi-optima. Experimental results show that all the parameters involved are calibrated with high accuracy.