视频图像的车辆检测与识别

摘要

A new method is proposed to extract candidate regions from video images.In this method,video image is converted into HSV space,the red areas of video images are extracted by H components and horizontal edge of vehicle's bottom by V components,then the candidate regions of vehicle can be determined by these two components.At the same time,optimized Gabor filters group is used to extract the characteristics of candidate regions, supported vector machine is used to train and recognize the selected characteristics.The filter group is improved by Quantum Evolutionary Algorithm(QEA),and niche cooperation evolution algorithm is introduced at this,also the improved filter group is clustered to reduce redundancy. Experiments show that this method is more accurate and faster,the improved Quantum Evolutionary Algorithm converges rapidly, so the optimal solution can be found quickly.%提出了一种新方法,用来提取视频图像中车辆的候选区域.即将视频图像转换到HSV空间,利用H分量提取图像中红色区域位置,V分量提取图像中车底的水平边缘位置,两者结合确定图像中车辆的候选区域.然后,利用改进的Gabor滤波器组对图像中的候选区域特性进行提取,最后利用支持向量机对提取的候选区域特性进行训练、识别.滤波器组通过量子进化算法进行了改进,其中引入了小生境协同进化算法并对优化后的滤波器组进行聚类减少多余的滤波器,降低冗余度.仿真结果表明此方法提取候选区域更加精确、快速.改进后的量子进化算法收敛速度快,能够快速地找到最优解.