摘要
针对复杂运动背景中慢速小目标检测误检率高,实时性差等问题,提出了基于自适应阈值分割的慢速小目标检测算法。首先计算连续两帧图像特征点的金字塔光流场,对光流场进行滤波,获取匹配特征点集合。然后对图像运动背景进行建模,拟合投影模型参数,通过投影模型得到运动背景补偿图像,进行图像差分处理,获得差分图像。最后迭代计算差分图像的自适应阈值,修正差分阈值,差分图像二值分割,检测出运动目标。实验结果表明算法能够准确地检测出复杂背景中的慢速小目标,虚警率为2%,目标漏检率为2.6%,目标检测准确率95.4%,每帧图像目标检测时间为38ms,能够满足运动目标检测对实时性的要求。
Aiming at the problem of high false detection rate and poor real-time performance in the background of complex moving background, a slow and dim target detection algorithm based on adaptive threshold segmentation is proposed. Firstly,the feature points of the two frames of the Pyramid optical flow field are calculated, and the optical flow field is filtered to obtain the matching feature points. Then the background of image motion is modeled, and the model parameters which are obtained by using the projection model is to get the motion background compensation images. Finally, The adaptive threshold of the difference image is calculated. The difference threshold is corrected, and the difference image is segmented, and the moving target is detected. Experimental results show that algorithm can accurately detect the complex background of slow and dim target, the false alarm rate is 2%, the undetected rate was 2.6%, target detection accuracy of 95.4%, 38 ms time for each frame of the video object detection can meet the requirements of real-time detection of moving targets.
出处
《电子设计工程》
2016年第6期77-80,84,共5页
Electronic Design Engineering
基金
国家自然科学基金(61471194)
航空科学基金(20155552050)
中国航天科技集团公司航天科技创新基金
南京航空航天大学研究生创新基地(实验室)开放基金(kfjj20151505)
中央高校基本科研业务费专项资金资助
关键词
目标检测
慢速
复杂运动背景
特征点
target detection
slow
complex moving background
feature points
