| Point set registration,which aims to establish correspondence between given pairwise and groupwise point sets,is a fundamental and critical problem in the field of computer vision and pattern recognition.At present,point set registration is widely used in medical image analysis,remote sensing image processing,and face recognition.Therefore,research on point set registration has high theoretic significance and practical values.In the thesis,a comprehensive literature review on non-rigid point set registration is presented.The state-of-the-art algorithms and modeling methods for non-rigid point set registration are given,with a summary of their pros and cons.Then,a non-rigid point set registration method incorporating local geometric structure constraints is proposed.Furthermore,in order to solve the problem of track-to-track association in the presence of missed detections and sensor bias in multi-sensor systems for intelligent vehicle,a novel track-to-track association method is proposed.The main contributions of this thesis are as follows.1.The more details of the state-of-the-art non-rigid point set registration methods are summarized.A comprehensive literature review on non-rigid point set registration is presented.The state-of-the-art algorithms and modeling methods for non-rigid point set registration are given,with a summary of their pros and cons.Then,some of the most prominent representative methods are selected to conduct qualitative and quantitative experiments.From the experiments conducted on different datasets,non-rigid point set registration by preserving global and local structures seems to outperform their rivals both in accuracy and computational complexity.2.A non-rigid point set registration method incorporating local geometric structure constraints is proposed.Using a Gaussian mixture model,the non-rigid point set registration problem is formulated as a maximum likelihood estimation problem in this thesis.Then two local structure descriptors are constructed to preserve the local structure of the same point set during the transformation.An expectation maximization algorithm is applied to estimate the unknown parameters.Therefore,non-rigid point set registration method incorporating local geometric structure constraints is proposed.Finally,the effectiveness of the proposed method is validated on different datasets.3.A track-to-track association method based on the point set registration for intelligent vehicles is established.The problem of independent track-to-track association for each frame in the presence of missed detections and sensor bias is considered.The idea of point set registration is applied to track-to-track association in this thesis.The local tracks of two sensors are considered as two sets of points.Gaussian mixture model is constructed based on the public track information estimated by two sensors.Then the track-to-track association with sensor bias is formulated as a maximum likelihood optimization problem with an expectation maximization algorithm being proposed to address it.Finally,the performance of the proposed method is verified by simulation and real intelligent vehicle experiments.In general,the non-rigid point set registration method proposed in this thesis has better performance than the state-of-the-art methods.In addition,A track-to-track association based on the point set registration algorithm for intelligent vehicles is proposed.The findings of this thesis will provide insights for the intelligent vehicles from both the theoretical and experimental perspectives. |
