论文部分内容阅读
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 generalized stochastic localization model and a Monte Carlo method for high noise environment are proposed. Based on the known georeferenced images, the precise location of the on-board panoramic image sequences is achieved. First, based on the Bayesian rule and Markov random chains, generalized stochastic models for global positioning of moving image sequences under both geometric and radial constraints are derived. Then, taking into account the 80% mismatch rate of multi-source images under difficult conditions such as shadows, occlusions and dynamic targets, an integrated Monte Carlo matching and localization algorithm is proposed based on particle filter theory. Through the prediction and updating iterative strategies, Eliminate the gross errors while getting the best positioning results. Through more than 2000 positioning experiments of panoramic video sequence, it verifies that the proposed method can overcome the problem that the traditional smoothing solution can not converge due to too many mismatched points in multi-source image matching.