运动背景中结合特征位移矢量场模糊分割与 OTSU法的运动检测

运动背景中的运动检测难度较大，背景运动补偿后差分以及分割光流场可实现动目标和背景的分离，差分前需进行鲁棒的背景估计，且差分后易出现空洞，而光流估计在噪声以及目标运动速度较大时并不准确，尤其在光照变化时，两种方法均易失效。本文提出一种特征点位移矢量场模糊分割与图像自适应阈值化相结合的运动检测方法，实现在无任何关于运动目标或者运动背景先验信息条件下的动目标检测。通过改进的 SIFT匹配方法生成鲁棒的特征位移矢量场，采用模糊 C均值聚类算法对 SIFT位移矢量场进行无监督分类，实现动目标与背景特征的自适应分离。 OTSU法和形态学操作实现图像的自适应分割，用以修正特征点凸包，最终分割出动目标区域。与鲁棒的背景运动补偿后差分以及光流估计的对比实验表明，在目标运动速度较大、光照变化以及噪声情况下，本文方法均能够检测出运动目标，且在光照变化下的优势明显。

Detection of Moving Object in Moving Background Based on Feature Vector Field Fuzzy Segmentation and OTSU Method

Motion detection in moving background is difficult. Background compensation followed by inter-frame difference and optical flow segmentation can separate object from background. Problems are that the former requiresrobust background estimation and may induce holes, while optical flow estimation is often invalid with noise, changes in illumination and high speed object. Two methods may be invalid especially when illumination changes. A new approach is presented based on feature displacement vector field fuzzy segmentation and OTSU method for motion detection in dynamic scenes without any prior information about object or dynamic scenes. A robust feature correspondences set is obtained by an improved matching strategy of SIFT, and Fuzzy C-means clustering algorithm is used to classify the generated feature displacement vectors. OTSU algorithm and morphological operations are performed for image thresholding, which modifies the convex hull of detected features. At last, moving object region is segmented. Compared with inter-frame difference and optical flow estimation, experiments demonstrate that the proposed method can detect moving object in the condition of noise, illumination change and object moving at a high speed.