中国数字人切片图像的自动配准与分割方法研究

提出了一种新的方法,用于解决数字人数据集重建前的切片图像配准和分割问题.首先利用改进的彩色空间动态阈值法和形态学处理提取切片图像的定位杆坐标,在此基础上对原始图像进行变换,从而实现数据集的配准;其次根据配准后数据集相邻切片目标轮廓在位置和形状上的相似性,利用前层切片分割结果指导初始化位置和限制搜索空间的主动轮廓模型,实现了配准后数据集的自动分割.所提出的方法在数字人切片图像的分割中取得了令人满意的效果,并且该方法还可以应用在CT、MRI等医学图像序列中的器官和部件的分割中.     

基于粒子群优化算法的多模态医学图像刚性配准

提出了一种基于轮廓特征点及利用PSO(粒子群优化)求解多模态医学图像自动配准新方法．首先采用数学形态学中腐蚀和膨胀算法对图像进行预处理,用区域生长法提取图像的边缘;再用subtractive聚类算法提取出轮廓特征点,将两个特征点集的均方根极小值作为配准准则,然后用PSO算法求解空间变换参数．该算法适用于多模态医学图像配准,与其他算法相比,PSO算法具有操作方便、可靠性好、不易陷入局部极值等优点。     

A design of space robot multi-target capture algorithm based on DSP

To correctly capture spatial targets from cluttered and motive celestial background, a new Multi-Target Capture algorithm was proposed, which is a comparative difference algorithm based on the combination of centroid extraction and despun registration of efficient points. Moreover, this algorithm was applied in an image processing system based on the DSP featuring high speed and high performance. The procedures of image processing are as follows: first, label efficient points in the frame and extract their centroids; second, make appropriate despun registration, according to the reference rotation angles provided by Space Robot position system; third, translate and register centroid coordinates of efficient points in reference frames and get the registration points according to the principle that there are the most same centroid coordinates of efficient points when completely registered; finally, eliminate the same background points by using comparative difference method. The result shows that this image processing system can meet the needs of the whole system.     

基于改进FPFH-ICP的车载激光雷达点云配准方法

为了改善传统车载激光雷达点云配准方法准确度低、计算速度慢的问题，提出了一种基于快速点特征直方图（fast point feature histograms, FPFH）初始匹配与改进迭代最近点（iterative closestpoint,ICP）精确配准相结合的改进FPFH-ICP配准算法。配准前使用体素滤波器和statistical-outlier-removal滤波器进行预处理；采用FPFH提取点云特征，基于采样一致性（sample consensus initial alignment, SAC-IA）进行初始配准，为精确配准提供良好的位姿信息；建立K-D树并在传统ICP配准算法的基础上添加法向量阈值，对车载激光雷达点云数据进行精确配准；在4种不同场景的实验中，改进FPFH-ICP配准比ICP配准的均方根误差和配准用时分别平均减少了7.56%和41.22%，比点特征直方图（point feature histograms, PFH）配准的均方根误差和配准用时分别平均减少了30.28%和18.95%，表明改进的FPFH-ICP能够对车载激光雷达点云数据实现精确且高效的配准。