基于小波系数及光学特征的车辆阴影检测

为消除投射阴影对检测交通实景图中行驶车辆的干扰，从阴影区域的图像特征入手，在对实景图与背景图实施“商比例”处理的基础上，提出了基于小波系数分布特征及阴影光学特征的车辆阴影滤除两步法．首先通过对高、低频带小波系数分析，提取出符合阴影特征的小波系数，重构运算后得到准阴影图；然后通过构造基于光学不变性的识别判据，成功识别出虚假阴影区．实验结果表明，本文方法对汽车对象的颜色、大小及阴影的投射方向没有特殊要求，阴影平均检出率超过92％，平均误检率低于3％，且速度较快；与汽车对象同画面行驶的自行车及行人阴影也受到有效抵制和滤除；在适用性、有效性和处理速度等方面反映出较优的品质．

Vehicle Shadow Detection Based on Wavelet Coefficients and Photometric Features

Casting shadow can cause serious disturbance to the correct detection of motion vehicles in the traffic current frame. To filter the vehicle shadow, a method based on the distribution characteristics of wavelet coefficients and photometric features of vehicle shadow was presented. First, according to the image traits of shadow region, the current frame and background frame was processed by ＂quotient-scaling＂. Then the low and high frequency wavelet coefficients were analyzed and the wavelet coefficients suitable for the shadow characteristics were extracted, from which, a quasi shadow graph was attained by operational restructure. The false shadow area could be identified by constructing a recognition criterion based on photometric invariance. The testing results show that the method has no strict desire to the color and size of the vehicle object, as well as casting direction of shadow. The average correct detection ratio of shadow exceeds 92 %, and the average false detection ratio is below 3 %, with high speed. At the same time, the shadows of bicycles and walking people in the same picture with automobiles are filtered efficiently. Therefore the method works well in the aspects of applicability, validity and operation speed