While the effect of traditional edge detection algorithms is largely dependent on the selection of threshold value, and a new edge detection method based on maximum local variation and 2D OTSU was proposed. The method calculated the difference of the gray values of the all pixels and the center pixel in a local area in an image and used the biggest difference values in different local areas to describe the edge distribution information of the image. To do this, an edge distribution information image could be obtained, and then the method used the 2D OTSU method to segment the edge distribution information image and gained the edge binary image. Based on the edge binary image and some prior information of the vehicle, an algorithm for car window area and driver area localiza-tion was proposed. In the end, a seat belt detection method was proposed by detecting whether there is a line meeting the seatbelt prior characteristics in the driver area. The experimental results show that the method can accurately locate the car window edges and the pi-lot area, and can be applied to detect the seat belts, and then has a certain practical value.%传统的边缘检测算法的效果很大程度上取决于阈值的选取,针对这个问题,提出了基于局部最大变化和二维OTSU的边缘检测方法,该方法利用图像局部区域的所有像素灰度值与中心像素灰度值的最大差值来描述图像边缘分布信息,从而得到图像边缘分布信息图,然后利用二维OTSU方法对该边缘分布信息图进行二值化处理得到边缘二值图。利用该边缘二值图,结合车辆的一些先验信息,提出车窗定位算法,并进一步确定驾驶员区域,最后通过在驾驶员区域内检测是否存在满足安全带先验特征的直线来判断驾驶员是否佩戴安全带。实验结果表明,该方法能够准确定位车窗边缘和驾驶员区域,可以应用于安全带的检测,具有一定的实用价值。
展开▼