自适应复杂背景干扰的运动目标检测算法  

Moving-target Detection Algorithm Adapting Complex Background Interference

作　　者：王东京 张宝辉[2] 陈弘原 王润宇[2] 吴杰 吴旭东 

机构地区：[1]南京理工大学电子工程与光电技术学院,江苏南京210094 [2]北方夜视科技集团有限公司南京研发中心,江苏南京211106

出　　处：《红外技术》2017年第11期1024-1031,共8页Infrared Technology

基　　金：云南省科技厅院士自由探索项目专项科研资助(2016HA009)

摘　　要：基于视频图像的运动目标检测,是根据目标的像素特征来判别出相对于背景运动的目标,当图像背景动态变化时,将难以区分背景和运动目标的像素特征,易造成检测错误。复杂背景下的运动目标检测是一大难点,目前主流的运动目标检测算法在背景灰暗、水面波动、气流颤动等复杂背景干扰下,难以准确地检测出运动目标。针对上述问题,提出一种自适应复杂背景干扰的运动目标检测算法,采用新的前景判断和背景模型更新方法,同时设计了一种创新型自适应阈值更新方法,当视频背景变化时,自动更新阈值。该算法增强了对复杂背景、镜头抖动的抗干扰能力,通过各种视频测试,背景点检测正确率达到0.9958,前景点检测正确率达到0.8012,极大提高了前景检测率,而且该算法满足高实时性要求,对复杂背景下的运动目标检测有显著效果。Video-based moving-object detection is according to the pixel characteristics of the target to determine the target that is moving relative to the background. When the background of the image changes dynamically, it will be difficult to distinguish the pixel characteristics between the background and the moving target, which is prone to detection error. Moving target detection in a complex background has been a challenging problem until now. The mainstream algorithm of moving target detection still cannot accurately detect moving targets under complex backgrounds such as dark background, water surface fluctuation, and airflow quiver. Aiming at the above problems, a novel moving-target detection algorithm was proposed for adapting to complex background interference. The algorithm adopted a new standard of classifying the foreground and a new method of updating the background model. At the same time, a creative method was put forward to update the threshold, which would automatically adjust the size of the threshold according to the change of background. The algorithm enhanced the ability to resist the interference of the complex background and camera shake. According to various video tests, the accuracy rate of background detection reached 0.9958.On the other hand, the accuracy rate of foreground detection reached 0.8012. Obviously, it is a great progress on foreground detection. Besides, the novel algorithm has high work-efficiency and has a significant effect on the extraction of moving objects in a complex background.

关 键 词：运动目标检测 复杂背景 模型更新 自适应阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]