鲁棒的PTZ摄像机目标跟踪算法

提出基于可旋转、变焦（PTZ）摄像机的目标实时智能跟踪算法.通过检测图像序列的方差,根据噪声和运动干扰的历史数据,建立相应的空间分布模型,实现有选择性的﹑高灵敏度的运动检测和云台控制.根据噪声的分布情况调整判决阈值,为不同区域赋予不同的检测灵敏度,提高了系统的噪声容限.自适应地调整方差的序列长度,实现系统灵敏度的大范围调整,以适应噪声、光照的快速变化和不同数量级别的运动幅度.推导长序列方差的快速算法,给出系统实时工作的例子.实验结果表明,算法工作可靠,响应迅速.

Algorithm of robust object tracking using PTZ camera

An efficient algorithm which robustly tracked a moving object in real-time using pan/tilt/zoom (PTZ) camera was proposed. Noise and interference were detected and stored in a spatial model after calculating the covariance of image sequence. Highly sensitive and selective motion detection of later frames was achieved by comparing later covariance with the spatial model. The detection threshold of each region was carefully measured to reduce noise and constant interference. The number of frames to calculate covariance automatically changed according to environment such as light, signal to noise ratio (SNR), and speed of the motion. An optimization method to calculate the covariance was proposed. Experimental results show that the algorithm worked robustly in various kinds of environments．