帧差法和Mean-shift相结合的运动目标自动检测与跟踪

传统的Mean-shift算法简单快速,但存在半自动跟踪缺陷,在起始帧需要手动确定搜索窗口来选择目标,且核窗宽固定不变,不能实时地适应目标尺寸大小变化,容易跟丢目标。接合帧差法,首先通过帧差法检测目标,并获取目标窗口和中心,再结合Mean-shift跟踪,并通过设定ρ^(y)相对改变量r来确定目标模板是否需要重新获取,实现Mean-shift算法全自动跟踪,并能适应目标尺寸大小改变的情况。实验表明,该方法跟踪准确,实时性高。

Moving target detection and tracking Based on Frame Difference Method combined Mean-shift

Traditional Mean-shift algorithm is simple and fast, but there are semi-automatic tracking defects, we need to determine the search window to select the target at the initial frame, and the bandwidth of kerne is fixed, not in real time to adapt to changes in target size, which is easily get lost during the tracking. This study apply the Frame difference method, first to detect the target and obtain the target window ,then to integrate Mean-shift tracking and to determine whether to obtain the new target template by setting the relative amount r of P(y) . Finally to achieve the Meam-shift algorithm automatic tracking, and adapt to the changing of target size. The experiments indicate that this method comes out with high accuracy in tracking and better real timing.