复杂密林地区植被点云组合滤波方法研究

Study on Combined Filtering Method of Vegetation Point Cloud in Complex Dense Forest Area

为提高密林地形激光雷达测量(Light Detection and Ranging,LiDAR)点云数据植被点和地面点精准分类精度,采用5大传统滤波方法对林区点云进行滤波精度评定,借助布料仿真滤波算法实施相应的初始滤波处理,经准确处理后得到地形特征相对完备的初始地表点,再对具有突出植被点的初始地表点实施迭代开运算,由此得出新的地表点。并借助Ⅰ类误差、Ⅱ类误差等一系列的指标获得科学的评估精确度。布料仿真算法与简单形态学算法组合滤波较传统滤波算法Ⅱ类、总误差均明显下降,Kappa系数呈增长趋势。结果表明,该方法适用于林区点云分类,可以获得较好的滤波效果。

In order to improve the accuracy of precise classification of vegetation points and ground points using LiDAR point cloud data in dense forest terrain,five traditional filtering methods were used to evaluate the filtering accuracy of forest area point cloud,and the corresponding initial filtering was carried out with the help of cloth simulation filtering algorithm.After accurate processing,the initial ground points with relatively complete terrain features were obtained.Then,the iterative opening calculation was carried out on the initial ground points with prominent vegetation points,and the new ground points were obtained.And with the aid of a series of indicators such as ClassⅠerror and ClassⅡerror,the accuracy of scientific evaluation was determined.Compared with the traditional filtering algorithm of ClassⅡ,the total error of the cloth simulation algorithm and simple morphology algorithm decreased significantly,and the Kappa coefficient showed an increasing trend.The results showed that this method was suitable for point cloud classification in forest area and can obtain better filtering effect.