基于分解的三维Otsu运动车辆检测方法

在户外交通环境下，采用传统目标检测方法容易产生大量噪声，传统三维最大类间方差(Otsu)方法能消除噪声，但是不能满足实时性。为此，提出一种基于分解的三维Otsu运动车辆检测方法。通过隔帧对称差分法得到2个差分图像，对这2个差分图像采用基于分解的三维Otsu法进行阈值分割，对这2个二值图像在使用数学形态学滤波后，求交集得到运动车辆目标。实验结果表明，该方法在视频帧出现不规则抖动的情况下，能实时和精确地检测出运动车辆。

Three-dimensional Otsu Moving Vehicle Detection Method Based on Decomposition

It is frequent to produce a large number of noise in the outdoor complex traffic environment by traditional object detection method. Traditional three-dimensional Otsu method can eliminate the effect of image noise, but can not satisfy the real-time demand. To solve the above problems, a method for three-dimensional Otsu moving vehicle detection based on decomposition is proposed. Two difference images are got by the method of interval frame symmetry differencing. Two binary images are obtained after using a method based on decomposition three-dimensional Otsu. Moving cars are got by seeking the common ground in these two binary images after filtering with mathematical morphology. Experimental results show that it can detect moving cars for real-time and accuracy in the case of jittering in the video frame.