摘要: 针对道路监控视频中特定车辆图像序列的关键帧提取问题,在运动对象检测的基础上,提出一种关键帧提取方法。将积分通道特征和面积特征作为图像特征描述子,结合AdaBoost训练分类器,实现道路监控视频车辆序列图像中关键帧的提取。通过运动对象前景检测技术获得出现在监控区域的运动车辆最小外接矩形图像序列,选择满足监控分析需求(车牌清晰度高,能判断车型)的若干帧作为正样本,其他不满足监控分析需求的作为负样本,提取样本图像的面积特征和积分通道特征,利用AdaBoost方法训练得到一个分类器,使用AdaBoost分类器对测试样本进行分类,根据打分规则提取关键帧。实验结果表明,该方法能提取运动车辆从进入到离开监控区域的序列图像帧中最清晰的图像,实现道路车辆监控视频分析数据的有效压缩。
关键词:
AdaBoost分类器,
积分通道特征,
道路车辆监控,
关键帧,
最小外接矩形,
运动对象检测
Abstract: An approach for key frame extraction based on foreground motional object detection is proposed,which aims at extracting the key frame of specific vehicle image series on road surveillance videos.This method achieves the extraction of key frame from series of vehicle images on road surveillance videos utilizing integral channel features and the area feature as the image feature descriptor,combined with AdaBoost classifier training.Minimum circumscribed rectangle images of moving vehicle appearing in road surveillance are obtained by foreground moving object detection.Images with clear vehicle license plates and car type which satisfy the demand for surveillance analysis are selected as positive samples while others are chosen for negative samples.It extracts the area feature and integral channel features and trains a classifier using AdaBoost.After testing samples are classified,it picks up key frames according to grading rule of AdaBoost classifier.Experimental results show that,for a series of vehicle images which begin counting when a motional vehicle enters into the surveillance area and end when it leaves,the algorithm presented in this paper effectively selects the most distinct and clearest image,realizing effective compression of video analytical data for road vehicle surveillance.
Key words:
AdaBoost classifier,
integral channel feature,
road vehicle surveillance,
key frame,
minimum circumscribed rectangle,
moving object detection
中图分类号:
袁晶,王炜,杨建,刘煜,张茂军. 一种道路车辆监控视频中的关键帧提取方法[J]. 计算机工程, 2014, 40(12): 282-286.
YUAN Jing,WANG Wei,YANG Jian,LIU Yu,ZHANG Maojun. A Key Frame Extraction Method for Road Vehicle Surveillance Video[J]. Computer Engineering, 2014, 40(12): 282-286.