Combining the optical flow technique and geometric active contour model to segment moving objects is a promising method. It not only can segment the contour of target accurately but also can be extended to the case with dynamic background. In order to combined these two techniques more effectively, two sides were improved: A new active contour model, which was based on motion edges, was proposed, so the impacts from low textured motion areas can be eliminated; reducing smoothing at those locations where edges in the flow field occur during the computation, the edge-enhancement nonlinear diffusion constraint was chosen as a smoothness term in optical flow computation. Then the flow field could preserve the motion discontinuities which were important information to segmentation, so a better segmentation result could be gotten. The experimental results show that this algorithm performs well, even in the situation of dynamic background. Compared with other methods, more accurate results can be gotten by this method.%将光流技术和几何活动轮廓模型结合起来对运动目标进行分割,一方面可以通过算法分割出运动目标轮廓,另一方面,使得算法面对背景变化的情况仍具有一定的适应性.为了更好地将二者结合起来,进行2个方面的改进;建立基于运动边缘的几何活动轮廓模型,从而克服光流计算中由弱纹理区域产生的运动空洞对分割的影响;引入边缘增强的非线性扩散作为光流计算模型中的平滑项约束,可以在平滑流场的同时减小运动边缘处平滑效果而保护边缘,进而提高运动目标分割的完整性.研究结果表明:与现有算法相比较,该算法能够更准确地分割出运动目标的轮廓,同时在一定程度上克服背景运动的情况.
展开▼
