一种通用随机定位模型及无模型约束观测值的粗差处理方法

Remote sensed mobile platform localization or georeferencing is a key concern in surveying and mapping, computer vision, robotics and intelligent navigation. However, each field only concentrates in a certain solving method, e.g., surveying and mapping prefers bundle adjustment and robotics prefers filtering. This preference makes solutions fail under some negative conditions in almost all fields. In addition, in solving process the existing gross error detection methods can only handle those observations having certain geometric constraints, e.g., co-linearity and polygonal, and cannot deal with gross errors from observations with no certain constraints. In this project we propose a generic probabilistic localization model, which integrates all kinds of solving methods, and can handle observations with arbitrary probability distributions rather than just with Gaussian distributions. We also present a new gross error processing method base on particle theory especially against gross errors in constraint-free observations. In detail, we look gross errors as a part of the entire probability distribution of observations and no longer as outliers; we deal the gross error processing as a part of localization models and no longer as separate modules, and obtain the optimum localization parameters with gross error elimination simultaneously. The proposed generic localization model and its solving methods will integrate, complement and promote studies in the several fields about platform localization. Also, the proposed gross error processing method fills the gaps in the research of gross error detection for constraint-free observations.

遥感运动平台定位在测绘、计算机视觉、机器人学等领域都是研究热点，但：各领域通常关注某一类解法，如测绘侧重光束法平差、机器人侧重滤波，导致各领域都有难以克服的难题。此外，在解算中现有粗差探测算法假设观测值必须具有几何约束模型，如共线条件、多项式，因此无法处理无模型约束观测值中的粗差。本项目提出一种通用随机定位模型，整合各类解法的优点，且能够处理任意分布的含噪观测值；针对无模型约束的粗差，提出一种基于Particle的粗差处理理论与方法，能够在高噪声环境下实现平台的精确定位。具体地，我们将粗差不再视作outlier，而是将其看作观测值概率分布的一部分；将粗差处理不再独立于定位算法，而是整合于通用定位模型中，在解算的同时实现粗差剔除。本项目提出的通用定位模型，将较好地整合相关领域的研究成果，实现互补和促进；提出的粗差处理方法，弥补了无约束型粗差处理的研究空白，将有望促进粗差理论研究上的一些进展。

遥感运动平台定位在测绘、计算机视觉、机器人学等领域都是研究热点，但：各领域通常关注某一类解法，如测绘侧重光束法平差、机器人侧重滤波，导致各领域都有难以克服的难题。此外，在解算中现有粗差探测算法假设观测值必须具有几何约束模型，如共线条件、多项式，因此无法处理无模型约束观测值中的粗差。本项目提出一种通用随机定位模型，整合各类解法的优点，且能够处理任意分布的含噪观测值；针对无模型约束的粗差，提出一种基于Particle的粗差处理理论与方法，能够在高噪声环境下实现平台的精确定位。具体地，我们将粗差不再视作outlier，而是将其看作观测值概率分布的一部分；将粗差处理不再独立于定位算法，而是整合于通用定位模型中，在解算的同时实现粗差剔除。本项目提出的通用定位模型，已经较好地整合了相关领域的研究成果；提出的粗差处理方法，弥补了无约束型粗差处理的研究空白。并发表SCI top期刊在内的相关论文10余篇。