面向森林高度提取的光学多角度立体观测影像模拟

Simulation of multi-view stereo optical imagery for extraction of forest canopy height

森林冠层高度是森林生态系统植被碳储量的重要指示因子之一。已有研究表明光学多角度立体观测(即摄影测量)数据具备森林冠层高度提取的潜力,然而光学影像容易受到云、雨等天气因素的影响,区域高精度森林冠层高度制图必需依靠多星多时相数据融合才能实现。现有星载摄影测量系统获取的影像分辨率和观测角度差异较大,系统研究影像分辨率和观测角度变化对多角度立体观测点云垂直分布的影响规律,是实现多星多时相数据融合的前提。森林光学多角度立体观测模型是开展该研究的有效理论工具,但现有“森林光学多角度立体观测模型(LandStereo)”仅具备沿轨前后视立体观测影像模拟能力,无法用于WorldView等侧视立体观测影像的模拟分析。因此,本文对LandStereo模型进行了改进,使模型具备林区任意指定观测角度的影像模拟能力,进而利用改进后的模型对正视、正侧视和斜侧视等典型观测角度影像进行了模拟,并对不同观测角度组合构成的立体像对森林高度的提取精度进行了分析。结果表明,改进后的模型能够用于刻画森林冠层多角度立体特征,观测角度的变化是利用立体像对提取森林冠层高度精度的重要影响因素。本文为系统研究林区任意角度组合的多角度立体观测点云随影像分辨率和观测角度的变化规律提供了理论工具。

Canopy height is one of important indicators of carbon storage in forest ecosystems.Previous studies have demonstrated that optical stereo imagery held great potential for deriving forest canopy height.However,optical images are easily affected by cloud coverage and rains.The regional mapping of forest canopy height has to be achieved by the synthesis of multi-sensor and multi-temporal imagery.The spatial resolutions and viewing angles of existing spaceborne stereoscopic systems are quite different.It is essential to make a systematic investigation about the impact of image spatial resolutions and viewing angles on the vertical distribution of stereoscopic point clouds,which is the basis for the synergy of images acquired at different seasons even by different cameras.The theoretical model for the simulation of optical stereo imagery over forested areas is necessary for such studies.The LandStereo is such kind of model,which can simulate the along-track(only In-track viewing) stereoscopic imagery like ALOS/PRISM and ZY-3.However,the current version of LandStereo has no mode to simulate the along-track(including In-track and Cross-track) stereoscopic imagery like Worldview-1/2/3.Therefore,this study reported the modification of the LandStereo model to have it being able to simulate images acquired by any possible viewing directions in forest areas.Firstly,the calculation method of linear array projection center coordinates is improved,from originally only considering the change of observation altitude angle to considering the change of azimuth angle and altitude angle.Secondly,Rigorous imaging geometric model is improved to obtain RPC of images acquired by any possible viewing directions.Based on the improved LandStereo model,the bare and mountainous forest images with altitude angle of 75° and azimuth angle of 0°,90°,and 225° are simulated to verify the accuracy and extract forest canopy height.The surface elevation extracted by the improved LandStereo model is consistent with the input DTM with high accuracy(r=0.99,RMSE=1.507),which proves the geometric accuracy of the improved LandStereo model.There are certain differences in the extraction accuracy of forest canopy height between stereo images of different angles.The results showed that the modified model could correctly simulate the stereoscopic features of forest canopy with given view direction,also initially demonstrated that view angle was an important factor affecting the estimation accuracy of forest canopy height by stereoscopic images.