提出了一种基于背景模型的针对建筑物的阴影检测及变化检测方法。传统的基于背景模型的目标检测算法认为影像局部区域的自然背景符合高斯正态分布,而含有人工目标的区域则不符合这种分布,从而将目标区与自然地物区区分开来。然而,这种背景模型不适用于中等比例尺的航空影像。本文通过对背景模型的改进,把自然地物和人工地物都视为背景,而把阴影视为检测目标,可以很好地实现建筑物的阴影检测,然后采用阴影补偿法来检测建筑物的变化。试验表明了本方法的有效性。

We present amethod for detecting shadows and changes ofman-made objects based on background\nmode.l Original object detection method based on background model assumes that mi age pixels of nature\nbackground can be modeled as a Gaussian distribution, however regions containingman-made objects do not\nmatch with this mode.l So objects can be detected from the nature background. Our tests proved this\nbackgroundmodel is not applicable tomedium scale aerial photos because in that scaleman-made objects take\nupmore percentages in the mi age and cannotbe reqarded as a tail inGaussian distribution. Thispaper mi proves\nthe backgroundmode,l and treats both nature andman-made objects as background, and shadows as detection\nobjects. Using thismode,l we firstly detect object’s shadows from anomaly ofGaussian distribution and use\nsome filter operators to elmi inate those shadows not from man-made objects. Our tests proved that the detection\neffect ofman-made object’s shadows is quite mi pressive. Then the original mi ageswere compensated using the\nshadow mi ages. Aftercompensating, we got mi ageswithoutshadows. At lastchange detectionwasmade base on\nthose compensating mi ages. For convenience of the expermi ent, mi ageswere partitioned into several areas of\ninterest(AOI), and coarse matching of counterpartAOIs of old and new mi ages was made. And then we\ndetectedman-made objects’shadows inAOIs, and those containingman-made objects are remained and those\nshadows detectedwillbe considered asman-made objects information and be discarded. Thenwe compensated\ntheAOIswith shadows detection mi ages, and used difference change detectionmethod to compare theseAOIs\nand got the final change detection results. Our tests used aerial mi ages taken at the year 2000 as old mi ages,\nwhose scale is 1∶8000, and aerial mi ages taken at the year 2002 as new mi ages, whose scale is 1∶15000.\nExpermi ent results prove ourmethod is effective.