摘要
针对夜间环境下基于摄像机的车辆检测方法存在精度低、稳定性差以及无法对车型进行有效识别等问题,提出一种基于Kinect深度虚拟线圈的夜间车辆检测与计数算法。首先对Kinect深度图像进行预处理,分别获得运动目标深度图(MDM)与空洞深度图(HDM)。然后在MDM与HDM上设置虚拟线圈,利用积分图像分别生成对应的一维运动信号,对其进行加权合成获得对车辆运动特征的表达,并在合成的运动信号范围内检测出车辆目标,并计算出车辆目标的几何特征,通过SVM对车型进行有效识别。实验结果表明,该算法对于单双车道的车辆计数正确率分别高达99.75%与99.25%,大小车型分类正确率可达99.80%,处理单张图片的平均时间仅为7ms。
Detection methods for vehicles based on video cameras have problems of low accuracy,poor robustness,and difficult to identify types of vehicles in nighttime situations.A method using virtual-loop sensors based on Kinect depth data is proposed for detecting vehicles in nighttime.Firstly,depth image from Kinect is pre-processed to derive the target Motion Depth Map(MDM)and the Hole Depth Map(HDM).Secondly,virtual-loop sensors are set on MDM and HDM respectively,and generate integral images to compute the one-dimensional motion signals.The motion signals from corresponding MDM and HDM are fused to formulate the description of vehicle motions,from which vehicles are detected and counted.Then geometric features of vehicles are extracted,and types of vehicles are recognized by using SVM.The results show that the proposed method can accurately detect and count vehicles in nighttime situations with recognition rates 99.75% and 99.25%for one-lane and two-lane scenarios respectively.Its classify accuracy is 99.80%in terms of distinguish light and heavy vehicles.The average time of processing one image is only 7 ms.
出处
《交通信息与安全》
CSCD
2017年第5期28-36,共9页
Journal of Transport Information and Safety
基金
国家自然科学基金项目(51679181,51208168)
湖北省科技创新专项重点项目(2016AAA007)
河北省普通高等学校青年拔尖人才计划项目(BJ2014013)资助
关键词
智能交通
夜间车流量检测
深度虚拟线圈
KinCCt
SVM
车型分类
intelligent transportation
vehicle detection in nighttime
virtual-loop sensors
Kinect
SVM
vehicle classification
