Currently

automated reconstruction of buildings from data acquired by airborne Li DAR has been an important research topic in photogrammetry

especially for the precise reconstruction of complex buildings. Given the various and complex structures of buildings and the discrete distribution of laser data

automatically reconstructing complex buildings using airborne Li DAR data is complex. Certain methods are only suitable for buildings with simple structure

whereas some methods reconstruct the building models with the aid of other data source

such as remote sensing images and ground plans. Hence

this study proposed an automated approach for complex shape building reconstruction based on key point detection that only makes use of airborne Li DAR point cloud to solve the problems on the automated reconstruction of complex buildings. The point clouds of different roof planes are extracted automatically by combining RANSAC segmentation and space segmentation. For each plane

the exact contour is picked up using the Alpha Shape algorithm. Public intersection line features are determined by topological relation of these planes

which will help correcting the initial key points. Finally

the precise building model is obtained. Detection of the key points is important for our method

which contains extraction of initial key points

judgment of topological relation of planes

correction of key points

and model reconstruction. During judgment of topological relation of planes

three pubic intersection line features should be determined

namely

public line segment

public radial

and public line. The final key points are then calculated correctly according to these restraints. Two different types of buildings and one urban area were chosen to verify our method. Building 1 is composed of many horizontal roofs with different height levels

without intersection among those roof planes; whereas Building 2 consists of some gable roofs with relative more complex relationship. The experimental results showed that RANSAC and space combined segmentation is fantastic

the initial key points present the sketchy structure of the building roof

the vectorized roof models show the public intersection line features reasonably

and the final reconstructed building models fit the remote sensing images well. To further validate the effectiveness of our approach

the same two buildings were processed using Terra Solid and the models were compared with our results. The comparison proved that the accuracy of our approach is close to that Terra Solid. The selected urban area contains flat roof buildings

gable roof buildings

L shape buildings

hipped buildings

and complex structure buildings c ombined with the above typical shapes. Successful reconstruction of this area proves that our approach is suitable to urban area point cloud with complex shape buildings and is effective and practicable. Our approach realized the automated reconstruction of complex shape building models by point cloud segmentation

contour extraction

and key point detection. Experimental results of different types of buildings and one urban area showed that:（ 1） Our method can reconstruct complex-shaped buildings effectively;（ 2） Compared with TerraSolid the reconstructed results

the accuracy of our method is close to that of Terra Solid and is valuable;（ 3） The method is restricted by point cloud segmentation. If the threshold in the segmentation is rigorous

some roof plane point clouds would be lost

which would have side effect on the final reconstruction results.