Abstract: At present, the height difference threshold determination method in 3D building change detection is highly subjective and poorly generalized. In this paper, an adaptive method for determining height difference thresholds based on normal distribution is proposed. Firstly, based on the theory that the difference between changed and unchanged image elements in 2D change detection obeys normal distribution, the method assumes that the height difference of 2-phase digital surface model(DSM) obeys or approximately obeys normal distribution, and analyzes the distribution of height difference between changed and unchanged regions. Then, the method of calculating the height difference threshold using expectation and standard deviation is proposed based on the Laida criterion. Finally, using sliding window to dynamically calculate the height difference threshold. In two experimental areas, the building change detection results based on the proposed method improve the completeness, correctness and detection quality by 7.2%, 1.5% and 7.8% on average, compared with the empirical threshold method. This indicates that the method has good generality.

Key words: normal distribution, building change detection, DSM, height difference threshold, Laida criterion