摘要
该文提出一种基于红外视频的车载行人车辆检测系统模型。利用混合高斯模型对梯形感兴趣区域提取前景,再根据前景进行目标的框定,对目标进行几何形状、灰度分布、运动速度、边缘梯度分布4类特征提取特征数据,并把特征数据代入事先用特征数据库训练好的神经网络对目标进行分类检测,以达到对行人车辆的检测判别并预警的效果。仿真结果表明,该检测系统算法运行速度快,虚警率低,可靠性强。
A model of traffic detection system is presented in this paper. Firstly, achieve the foreground extraction within the region of interest（ROI） based on Gaussian mixture model（ GMM）, and make a presumption of valid target according to the foreground. Then extract the characteristic values of the valid target on the basis of its geometrical shape, gray-scale distribution, movement speed, edge gradient distribution. And detect these values by putting the characteristics date into the well-trained BP neutral network, so that thedetection, discrimination and warning of the traffic can be realized.
出处
《杭州电子科技大学学报（自然科学版）》
2013年第5期106-109,共4页
Journal of Hangzhou Dianzi University：Natural Sciences
基金
国家级大学生创新训练资助项目(GJ201210336014)
关键词
梯形感兴趣区域
混合高斯模型
图像特征值
神经网络
region of interest
Gaussian mixture model
image feature extraction
neural network
