基于光流场与水平集的运动目标检测

利用光流场信息及运动内极限约束确定运动目标区域的初始分割，提取光流大小与光流方向两个特征构成特征向量，使用K-means聚类算法获得运动目标区域，利用水平集方法对初始运动区域进行进一步分割，通过最小化定义的能量函数驱动代表运动目标的闭合曲线进行演化，曲线演化将在空间梯度较大的位置停止，从而得到运动目标的封闭边缘曲线。实验表明，该方法可有效地从图像序列中检测出完整的运动目标。

Motion object detection based on optical flow field and level set

Information of optical flow and epipolar constraint were used to get the initial segmentation region of motion object. The value of motion speed and the angle value of optical flow were extracted and feature vector was constructed using these two features. K-means cluster algorithm was used to get the region of motion object and level set was used to get information of image spatial segmentation. Minimization of the energy function led to the optimal segmentation of moving object by curve evolution. Evolution curve stopped at the position that spatial gradient was great and the close curve of moving object was produced. The experimental results show that the method can detect the whole motion object from image sequence.