在实时的车型识别系统中,由于光照的影响,需要一种简单快速有效的方法将汽车车体与其阴影分割开。利用阴影的光谱属性,同时根据阴影的几何特征及阴影区域内的点和汽车的空间位置、形状等相关特点,提出一种基于小波变换多分辨率特性的阴影分割算法,该算法能有效地分割出阴影与目标之间的分界线。先利用阴影的光谱属性对阴影进行粗分割,然后利用小波变换的多尺度性对候选阴影点进行特征提取,从而获取最后的阴影分界线。该方法不需要事先确定光源的方向、车体的彩色信息和背景纹理信息,能有效地分割任何颜色、任何背景纹理下、任何光源方向下的运动汽车投影阴影。实验表明,该方法定位准确,处理速度快,抗噪能力强,为进一步的车型识别提供基础。

In real-time vehicle type recognition, shadows interfere with moving vehicle extraction, location and recognition due to light. We propose a simple and effective method to segment moving vehicle cast shadow. The proposed method exploits spectral and geometrical properties of shadows and relationship between the point in shadow region and space position and shape of vehicle. Feature points of occluding function are detected using multi-resolution of wavelet transform and the boundary between self-shadow and cast shadow is detected. Firstly, the shadow can be coarsely segmented by the shadow spectral property, and then the feature points are extracted by multi-scale wavelet transform. The proposed method can effectively segment moving vehicle cast shadow. It needs not know in advance the information of the light direction, vehicle color and background texture. Our experimental results show that the proposed method is robust and effective in detecting shadows.