提出一种通用随机定位模型及针对高噪声环境的蒙特卡罗解法，基于已知地理参考影像实现地面车载全景影像序列的精确定位。首先，基于贝叶斯准则和马尔可夫随机链，推导了几何、辐射两种约束条件下运动影像序列全局定位的通用随机模型。然后，顾及阴影、遮挡、动态目标等困难条件下的多源影像匹配80％的误匹配率，基于粒子滤波原理提出蒙特卡罗匹配与定位一体化求解算法，通过预测、更新的迭代策略，在剔除粗差的同时获得最佳定位结果。通过2000余张车载全景影像序列的定位试验，验证了本方法能够克服多源影像匹配中误匹配点太多导致的传统平差解法无法收敛的问题。%A generic probabilistic localization model and corresponding Monte‐Carlo solutions especially designed for high‐noise environments are presented to georegistrate a ground mobile vehicle with mounted panoramic camera to geo‐referenced ortho‐images .Firstly ,the probabilistic localization model is deduced according to Bayes rules and Markov chain under the two constraints of geometry and radiance .Then a particle filtering method called Monte‐Carlo is introduced to solve the localization model , considering the difficulties of multi‐source matching between panoramic images and ortho‐images caused by shadows , occlusions ,moving objects etc .,and achieves the matching and geo‐referencing simultaneously .Tests with more than 2000 panoramic images and one ortho‐image with 0 2.5 m accuracy proved that our method can tolerate excessive blunders more than 80% caused by mismatching ,and demonstrated obvious advantages over traditional bundle adjustments that highly sensitive to gross errors .
展开▼
