摘要
图像运动目标跟踪是自动监控、道路导航、交通监控等重要系统的核心技术。为了解决视频中多个运动目标的跟踪,利用帧间差分、Kalman滤波器以及少量特征匹配来自动提取各个运动目标并进行跟踪。针对灰度图像序列,采用一种基于图像序列的梯度信息,使用帧间差分法自动提取运动目标,利用Kalman滤波器预测目标位置,在估计目标在下一帧的位置范围后,根据目标直方图特征来缩小搜索范围实现目标对象的准确跟踪。实验结果表示,该方法具有较小的运算量和较好的实时性,同等条件下具有较高准确性。
Image motion tracking is the main technology of automatic visual surveillance, pathfinding, traffic monitoring and other important systems. Uses conterminous frame differing, Kalman filter and feature matching to detect multi-target in the video frequency. Uses conterminous frame differing based on gradient to detect multi-target. Realizes target tracking by the histograms after the target location in next frame is obtained by Kalman filter. The experiment indicates that this algorithm is of less computation, stronger anti-interference and higher accuracy.
基金
公安部科技创新项目(No.2010YYCXGDST066)
江苏大学本科生科研立项(No.10A124)