Building Detection from LiDAR and Optical Data Using Support Vector Machine in Pixel-Based and Object-Based Analysis Print

Building detection from areal and satellite images is an active discussion in remote sensing and machine vision in recent years. Urban areas usually are dense and consist of complex components such as compact tree areas and buildings with gable roof and glassy parts. Classification algorithms which are applied to these kinds of data sets will be faced many problems. In this paper to deal with the aforementioned problems, the object based features; height and etc. have been investigated for classification by the use of support vector machine in the object based and pixel based analysis. It should be noted that pixel based analysis performed in two different states with features which are extracted from aerial imagery and LiDAR data. The proposed method consists of three general steps; the first step is data preparation and features extraction. The second step is classification by the use of support vector machine in object based and pixel based analysis; In the third step, post processing is applied then results of classifications are compared and evaluated with ground truth data. In this study the final goal is to achieve optimized algorithm using various features. with comparison of Kappa coefficient in three classifications; o.97 in object based analysis, o.88 in first state of pixel based analysis and 0.97 in second state of pixel based analysis, it is obvious object based analysis achieved the best result due to using features such as shape and structure. More over using LIDAR data in second state of pixel based analysis increased the accuracy of pixel based classification.