基于对称帧差和分块背景建模的无人机视频车辆自动检测

发布时间：2018-05-04 10:37

本文选题：智能交通 + 车辆检测　；参考：《东南大学学报(自然科学版)》2017年04期

【摘要】：为了从广域的视角准确全面地识别交通流信息,针对无人机视频提出了基于对称帧差和分块背景建模的车辆自动检测方法.首先,对视频图像进行4×4降维处理和灰度化处理,并人工勾勒出感兴趣区域(ROI),以降低图像维度,划定检测区域;其次,利用对称帧间差分法提取ROI中的运动目标,并在此基础上应用分块背景建模获得背景图像;然后,通过背景差分初步提取车辆信息;最后,基于形态学处理等方法消除噪声,实现车辆识别.此外,提出了针对车辆识别算法的正检率、重检率、漏检率和错检率4个评价指标.基于150帧无人机视频图像对算法进行测试,结果表明:算法具有较高的正检率(均值92.29%)、较低的漏检率(均值7.31%)与错检率(均值0.39%),而重检率为0.
[Abstract]:In order to accurately and comprehensively identify traffic flow information from a wide area perspective, an automatic vehicle detection method based on symmetrical frame difference and block background modeling is proposed for UAV video. Firstly, the video image is processed with 4 脳 4 dimensionality reduction and grayscale processing, and the region of interest is manually delineated to reduce the image dimension and delimit the detection area. Secondly, the moving object in ROI is extracted by symmetrical inter-frame difference method. On this basis, the background image is obtained by using block background modeling. Then, the vehicle information is extracted by background difference. Finally, the noise is eliminated based on morphological processing to realize vehicle recognition. In addition, four evaluation indexes for vehicle recognition algorithm, such as positive check rate, recheck rate, missed detection rate and error detection rate, are proposed. The algorithm is tested based on 150 frames of UAV video images. The results show that the algorithm has higher positive detection rate (mean 92.29), lower miss rate (mean 7.31) and false detection rate (mean 0.39%, and recheck rate 0).
【作者单位】：重庆交通大学交通运输学院;
【基金】：重庆市社会事业与民生保障科技创新专项资助项目(cstc2015shms-ztzx30002,cstc2015shms-ztzx0127) 重庆市教委科学研究资助项目(KJ1600513) 重庆交通大学科研启动资助项目(15JDKJC-A002) 重庆市科委基础科学与前沿技术研究资助项目(cstc2017jcyjAX0473)
【分类号】：TP391.41

【相似文献】