着陆避障中双目立体匹配方法与仿真

摘要： 在着陆避障过程中,利用双目相机拍得的图片进行立体匹配得到稠密的地形高程图是一种有效的方式.文中对绝对误差之和(sum of absolute differences, SAD)、全Census变换以及稀疏Census变换立体匹配方法原理进行说明,针对SAD算法无法处理左右图像亮度不一致的情况提出了对匹配窗口进行灰度规范化以及加入LOG滤波的改进措施,并且对上述方法分别应用数据库图像和沙盘拍图进行仿真测试及比较,从匹配错误率和实时性的角度分析各算法在地外星体着陆这种特殊场景中的使用性能.     

融合明度特征的V-SAD立体匹配算法

针对SAD算法难以对图像中纯色与弱纹理部分进行准确匹配的问题,提出了将HSV空间明度特征与SAD算法相融合的立体匹配算法,称为V-SAD算法.首先将图像从RGB空间转化至HSV空间,并根据H、S、V值将像素点按照颜色分为10类,同时得到明度特征图.然后结合SAD算法需要的灰度特征图计算匹配代价,得到初步的视差图.接着,...     

Best practices for model-based and simulation-aided engineering of power transmission and motion control systems

This paper deals with the modelling and simulation of aircraft systems, in particular for power transmission and control. It is intended to review, propose and disseminate best practices for making model-based/simulation-aided engineering more efficient at any phase of the system life cycle. The proposals are aimed at creating value, not only by increasing the performance of the product under study but also by shortening the time to market, capitalizing knowledge, mitigating risks and facilitating concurrent engineering. The needs associated with the engineering activities are firstly identified to define a set of requirements for the models. Then, these requirements are used to drive the considerations leading to model development, focusing in particular on the process, modelled physical effects, modelling level, model architecting and concurrent engineering. The third part deals with the model implementation, giving special consideration to the different types of models, causalities, parameterization, implementation and verification. Each part is illustrated by examples related to safety critical actuators.     

特征融合的双目立体匹配算法加速研究与实现CSCD

随着图像分辨率和场景信息获取实时性需求的提高,业界对双目立体匹配算法的效率提出了更高的要求。针对该问题,提出了将SAD与Census变换特征融合的结果作为初始匹配代价,利用SGM算法进行代价聚合,采用赢家通吃策略计算视差,通过左右一致性检验检测出遮挡点并填充,使用中值滤波剔除异常值,最终获取优化后的视差图。采用统一计算设备架构(CUDA)对算法实现并行计算,针对立体匹配比较耗时的问题,该算法最大化地利用共享内存、寄存器内存以及CUDA流,实现了不同核函数之间的并行,大大提升了执行效率。结果表明,该算法在Middlebury立体匹配平台上,平均误匹配率下降了8.05%;在NVIDIA GeForce GTX 1650平台上运行450×375分辨率的图像,比原始SGM算法快687倍,运行高分辨率图像时依然能够实现实时显示性能。