基于Snake技术的运动目标轮廓提取

Locomotory Objects Detecting Based on Snake

提出了一种利用改进的Snake技术检测视频序列中活动目标的方法,首先改进内能项,用控制点之间的距离平方和作为弹性能量项以取代传统的长度,并构造局部能量窗搜索最优解,提高了Snake收敛速度。通过改进外部能量项,引入梯度矢量流算子,使Snake能够较好地收敛到目标的凹形边缘。最后对控制点初始位置、采样密度等影响收敛效果因素的选定作了适当的改进,给出了一种在Snake运动中动态调整其形态以提高收敛效果的策略,使Snake曲线能更快速、更准确地拟合活动物体的真实轮廓。实验证明,该算法能对视频序列图像中的活动目标轮廓进行较好的提取。

This paper lays emphasis on the unsolved problem. First, improves the internal energy computation and uses the sum of the squares of the distances between adjacent control points instead of the length in the elastic energy which is a part of the internal energy. And it could see that the Snake is better behaved. Then, the external energy is also improved. The paper introduces the algorithm of gradient vector flow to settle the problem of tracking the concavity edge. At last, it departs the process of computing the Snake contour into several phases, in the processes, ituses different sample interval of the control points and variable coefficients to make Snake contour more close to the real edge. And besides, discusses the effectiveness on the Snake contour of the density of the control points. The experimental results comparing with traditional Snake contour prove that the algorithm has a better performance in tracing object＇s contour.