一种高精度视频目标检测与分割新方法

根据动态视频场景与多目标检测的应用需求,提出了一种变分光流场与mean shift图像分割相结合的高精度运动目标检测与分割新方法.依据动态场景多运动目标检测的约束条件提出变分光流场优化计算模型,并给出其数值解法.在此基础上提出结合meanshift的高精度运动目标检测与分割算法,此方法对摄像机运动和静止情况都适合,能够进行同一场景中多个运动目标的高精度检测,并且不需要事先的学习和人工干预,具有通用性.

Novel approach to video object detection and precise segmentation

Taking into account multiple objects detection in dynamic video scene, a method based on variational optimization of optical flow and mean shift was proposed. A variational optimization model for optical flow field was built according to constraints of multiple objects detection in dynamic video scene. The proper numerical solution was then given by this kind of optimizing calculus of variation. With the optical flow optimized solution, a high precise segmentation method was proposed. This approach can be used in the situations of both motion camera and stationary one. Meanwhile it can also be used to detect several targets in dynamic scenes simultaneously without learning in advance and manual intervention so as to be implemented automatically. A lot of experiment results validate the effectiveness of the proposed method.