序列图像中运动目标检测

提出动态背景下序列图像中的运动目标检测算法。利用像素邻域的各向同性对图像进行归一化,消除亮度变化等因素的影响;利用光流信息并结合小波变换由粗及精计算速度场来配准图像;用当前帧作参考图像,通过时域积分校正背景图像。当前帧与校正后背景图像作差得到差分图像。假设该差分图像中噪声分布为高斯分布,由高斯分布的3σ特性滤除差分图像中的噪声,则粗定位出目标;最后以聚类方法确定运动目标区域。分别对200帧可见光和200帧红外图像序列进行实验,检测率分别为95%和94%。

Detection of moving targets in video sequences

A method for detecting moving targets is proposed, which combines optical flow and wavelet transform, through using spatial and temporal information in image sequences under condition of moving camera. Image is normalized to remove influence from illumination variation based on isotropy in neighbor of pixel. The velocity of background is estimated by optical flow, affine motion model and wavelet pyramid structure. Background of image is corrected by using the current frame for motion compensation based on temporal integration. The current frame subtracts the warping background and filters some gauss noises by Gaussian distribution. Finally, the region of moving object is decided by a cluster method that is proposed to adapt to this algorithm. Both visual optical images sequence of 200 frames and infrared images sequence of 200 frames are experienced respectively. The experimental results show that the detection ratios reach 95% and 94% respectively.