基于改进Otsu算法在前方目标车辆识别中的研究

摘要

In the anti-collision warning system,the common image segmentation algorithm in identifying vehicle often causes the phe-nomenon that loss of information about the target feature and edge blurring. Therefore,it puts forward an improved image segmentation algorithm based on the Otsu algorithm and GA in this paper. This algorithm first confirms a threshold range of the front lane area by the normal distribution function,then crossovers and mutates the initial population of GA to search the optimal segmentation threshold. On this basis,combined with the adaptivity of Otsu,it searches the local optimum in the range of a set threshold. Finally,the optimal thresh-old of image segmentation is obtained. The results show that the improved algorithm has a good recognition performance in the light of the dim light and the damage of the camera. It not only can effectively segment the target vehicle and the background,but also greatly shorten the running time and improve the efficiency of the segmentation. Through simulation,the improved algorithm has obvious advan-tages in identifying the target vehicle in front.%在防碰撞预警系统中，使用常见的图像分割算法识别车辆时，易造成目标的特征信息丢失以及边缘模糊化的现象，因此文中采用了一种基于最大类间方差法与遗传算法相结合的分割算法。该算法先用正态分布概率密度函数确定前方车道区域灰度值的阈值范围；然后对遗传算法的初始种群进行交叉、变异操作以寻找分割阈值的最优解；在此基础上，结合最大类间方差法的自适应性，在给定的阈值范围内进行局部最优搜索；最后，获取图像分割的最佳阈值。研究结果表明，在光线昏暗、摄像头破损等情况下，改进算法具有很好的识别性，它不仅能有效地分割出目标车辆与背景，同时还大大地缩短了运行时间，提高了分割的时效性。通过Matlab仿真实验得出，改进算法在前方目标车辆识别中有明显的优势。