光电跟踪中实时视频对象分割算法

An Algorithm of Video Object Segmentation in Real-Time in Electric-Optic Tracking

光电跟踪系统研究中难点的问题是实时跟踪目标。针对光电跟踪系统所跟踪的空中目标在视频序列中的特性,将整个跟踪过程划分为远距离段、航前(后)段、航捷段。然后对远距离航段,采取一种弱小目标的动态规划检测算法,及时发现并跟踪目标;在航前(后)段,采取一种基于自适应双波门的视频对象分割算法分割出运动目标,计算出目标形心与跟踪物镜的角偏差,以偏差作为光电跟踪伺服系统的输入信号,驱动跟踪器跟踪目标;在航捷段,将每帧视频变换为一组分辨率从高到低类似金字塔式的分层框架,对目标采取从粗到细的方式进行分割,以实现视频对象占据大部分视窗时的目标快速、准确分割。通过实例验证了所提出的算法的有效性。

One of the hotspot and difficult subjects in electric-optic tracking systems is to segment video object accurately in real time. First,under the consideration of the characteristics of targets of electric-optic tracking systems in video sequences,the whole tracking process is divided into the remote section of track,the front （rear） of track and the course short of track. Second,a dynamic programming algorithm is used to detect small targets in the remote section of track,which can help electric-optic tracking systems to find and track small targets. Then,on the basis of adaptive double windows,an algorithm of video object segmentation is used to segment motion targets in the front （rear） of track and the angle deviations between the centroid of targets and the tracking lens,which are the input signals of the servo control system of electric-optic tracking system,and can be computed and used to drive the tracking device to track the target. In the course short of track,each video frame is transferred into a group of frames from high resolution to low resolution,like a pyramid layered framework. And then the video object is segmented in terms of an approach from rough segmentation to delicate segmentation such that the object can be fast and accurately segmented when video object occupies a majority of the screen of electric-optic tracking system. Finally,experiment simulation results demonstrate the valid of the proposed algorithm.