基于改进水平集的多运动目标检测方法

Method of Detecting Multiple Moving Object Based on Improved Level Set

针对多目标检测中目标和阴影连接导致检测失败的问题,提出一种基于改进水平集的精确检测方法.通过对监控视频进行对称差分获取运动信息,将其融合到水平集演化函数中,然后进行一次分割得到演化曲线,以此获得目标的位置信息.根据目标与阴影的差异信息,调整水平集演化函数,进行二次分割可准确分离多个目标及其阴影.将文中方法应用在公路监控的多运动车辆检测实验中,能够有效地检测出运动车辆轮廓,处理速度为24帧/s,消除阴影的准确率达98.6%,取得了准确的检测效果.

In the case of multiple moving object detection, the join of objects and shadow always leads to the failure of object detection. To solve this problem, an improved level set method is proposed. Moving information is extracted via symmetrical difference and it is merged into the velocity function of level set. The evolved curve is obtained after the first segmentation, hence the information of object location could be found. According to the information difference between object and shadow, the velocity function of lever set should be adjusted or modified. After the second segmentation, the multiple objects and their shadow can be separated accurately. The proposed method has been tested to detect moving object in motorway monitoring system. Experimental results show that the proposed algorithm is able to extract the contour of moving object effectively. Its processing speed is 24 frame/s and accuracy rate of shadow elimination reaches 98.6 %.