Urban land cover classification of oblique aerial imagery using object-based image analysis method

By means of airborne multiple camera imaging system, we can acquire vertical and oblique aerial images (VAI and OAI) at the same time. In addition to the reduction of data cost, the OAI can also strengthen the imaging geometry during aerial triangulation and be applied on automatic façade texture mapping. With the development of image matching technique, instead of airborne laser scanning (ALS), we can obtain surface point clouds by dense matching through both VAIs and OAIs. Comparing to the ALS data that were affected by the laser scanning angle, the photogrammetric points can provide much more information on the façade of buildings since the given information from the OAI. Therefore, the use of OAI in building verification and detection, 3D GIS, digital maps or other cyber-city related applications. In this study, we perform image classification using the original oblique aerial imagery and object-based image analysis (OBIA) method. We classify the OAI into six classes namely tree, grass, façade, roof, road and others. In OBIA, we utilize the multiresolution segmentation algorithm to separate the image into objects by merging pixels with similar color and shape homogeneity. Then, the objects are classified by different features such as color, shape, texture and object related features. In our study, we also use the "height map" and "gradient map" generated by back projecting the dense matched point clouds to the OAI to assist for urban object detection. The classification result shows that we can differentiate façade and roof from buildings successfully with the assistant of the height and gradient information. In the meanwhile, the classification result can further offer the semantic information from the OAI to 3D building models.