基于灰度相关的帧间差分和背景差分相融合的实时目标检测

Real-time target detection based on combining gray correlation frames differencing and background differencing

提出了基于灰度相关性帧间差分和背景差分相融合的一种实时目标检测方法。首先,采用基于最小均方误差的灰度相关性帧间差分,考虑图像噪声的高斯模型,确定差分用的阈值集合,得到差分二值图像;然后,使用混合高斯模型(GMM)进行背景建模,将模型的2.5倍方差和均值的最大值作为匹配准则,并基于光照变化对其背景图像进行自适应更新;最后,在原彩色图像的HSV空间中检测阴影,对帧间差和背景差后的二值图像去除阴影,接着进行相与操作,最终准确检测出运动目标。实验和分析结果表明,本算法能够消除噪音、阴影,能有效适应不同光照条件下的检测场景,因此,具有较高的准确性、鲁棒性和自适应性。

A novel real-time target detection algorithm based on combining gray correlation frame difference and background difference was presented. Firstly, the minimum mean square error of the gray correlation frame difference was used to get the difference binary image, and the difference image threshold was set considering the Gaussian model of the image noise. Secondly, background is modeled by Gaussian mixture model (GMM) using 2.5 times of the model variance and the maximum mean as matching criteria, and background was updated adaptively based on the illumination change. Thirdly shadows in the HSV space of the original color images were detected and shadows in the binary images of image difference and background difference were removed respectively, and logical AND operation was done. Finally, the moving target was detected accurately. The analyses and experiments results show that the proposed algorithm can eliminate noise, shadow, and can adapt to different illumination condition detection scenes. Therefore, it has higher accuracy, robustness and adaptability.