基于水平集运动目标检测与跟踪方法

Moving Object Detection and Tracking Using Level Set

针对目前常采用的运动分析检测方法存在的缺点,提出了基于水平集理论的测地线活动轮廓模型与背景差分相结合的运动目标检测方法.该方法使水平集函数免重新初始化,大大减少了曲线演化迭代的次数和运行时间,得到准确的运动目标轮廓.通过与粒子滤波和mean shift跟踪方法的比较,最终采用效率最高、最优的Kalman滤波预测物体的运动轨迹.实验结果表明,该方法对刚性和非刚性两类目标都具有较好的检测与跟踪效果.

According to the disadvantages of the moving detection methods which were used commouly,the moving object detection method of combination of geodesics active contour model and background subtraction based on the principle of level set proposed.The level set function is used in the improved method without re-initialization,the number of iteration in the curve evolution and the execution time are significantly reduced,the exact moving object contour is obtained.Through comparison with the particle filtering and mean shift tracking methods,finally the highest efficiency and optimal Kalman filtering was used to predict the moving trajectory of objects.Experimental results showed that the method has better detection and tracking effect on the two kinds of rigid and non-rigid objects.