复杂背景下单个运动物体的实时视觉追踪算法

介绍了一种完全基于图像信息的运动检测视觉追踪算法.该算法综合了已有的一些算法,并在他们的基础上进行改进,可以实现在摄像机运动不剧烈的情况下,对单个运动物体的追踪.由人工选取目标,计算机开始在第1帧和第2帧图像中提取特征点,并在两帧图像中对特征点进行匹配.利用匹配的特征点建立仿射运动模型,以估计背景的运动和预测目标位置.假设运动目标所占的像素面积很小,在预测点附近的一块小邻域内进行光流分割得到运动目标.该算法在640×480大小的两帧连续图片上验证,取得了较好的效果.在PC上的实验证明,设置适当的参数,本算法可以应用于12.5Hz或25Hz的图像采集频率.

Real-time visual tracking algorithm for single moving object in clustered environment

A visual tracking algorithm was introduced, which was completely based on image information. Several existing algorithms were integrated and improvements were made to achieve the goal of tracking one single moving object when the camera-s movements were small. A target was manually selected. Then feature points were extracted from the first and second frames and feature matching was conducted between the two frames. Using the matched features, affine motion model was established to estimate background movement and target position. With the assumption that target occupies only a few pixels, the separation of optical flow was calculated in a fixed small area around estimated target point and finally the moving object was obtained. The algorithm was applied on two consecutive images of the size of 640×480, and performed well. The experiments on PC show that with proper parameters the algorithm is able to work in the sampling frequency of 12.5Hz or 25Hz.