基于梯度统计和区域生长融合的运动车辆阴影检测方法

Moving Vehicle Shadow Detection Approach Based on Region Growing and Gradient Statistics

针对在视频对象分割时,运动阴影常被误分为视频对象的问题,提出了一种融合阴影模型和特征的阴影检测方法。用聚类识别的算法提取背景,获得包含阴影的前景,算出前景与背景梯度差。对梯度差投影,根据投影后得到的序列及其差分序列进行阴影区域初步检测,根据初步检测结果,运用区域生长算法搜索出全部阴影。实验结果证明,本方法检测效果好,速度快,能够应用于车辆阴影的实时检测。

To prevent the moving shadow from being misunderstood as a part of moving objects in video target segmentation, a novel approach is proposed for the shadow detection, in which the model and the characteristics of the shadow are merged. Using the clustering algorithm to extracte the background, then obtain the foreground with the the shadow. Calculate the gradient difference of the foreground and the background, and carry out the projection of the gradient, according to the sequence and the difference sequence after the projection identify the region of the shadow, and search the shadow using the regional growth method based on the initial result of the shadow detection. Experimental results show that the algorithm has a good performance, and requires fewer time. It can be used in the real-time detection of the vehicle shadow.