基于光流法的篮球图像运动块差异自主检测方法

为了解决传统方法在人群密集遮挡情况下无法满足检测要求,以及针对正常运动块的加速度改变,容易误检测为差异运动块的问题,通过光流法研究篮球图像运动块差异自主检测问题。分析光流法方程,通过篮球图像序列中像素强度数据的时域改变与相关性判断运动块像素的变化。通过码本模型对篮球图像前景位置进行提取,避免人群遮挡干扰,在前景位置处找到特征点。获取目标特征点后,通过光流法对运动块进行跟踪,针对全部能够被跟踪的运动块,引入光流运动方向数据;把光流运动方向在相同角度区间中的特征点当成一组数据完成归一化处理,降低对正常运动块的误判断,增强检测精度。对各区间中光流加速度进行高斯滤波处理,把各角度区间加速度累加,将其当成篮球图像加速度,设定累积加速度阈值,在图像块累积加速度高于设定阈值的情况下,认为出现差异情况。结果表明,所提方法能够检测遮挡背景下篮球图像运动块差异,NMI与LODF值均较大。说明所提方法满足遮挡情况下运动块差异检测要求,检测结果准确,不容易出现误检测现象。

Autonomous Detection of Motion Block Difference in Basketball Image Based on Optical Flow Method

In order to solve the problem that traditional methods can not meet the detection requirements under dense crowd occlusion, and that the acceleration of normal motion blocks changes, which is easy to be misdetected as differential motion blocks, this paper studies the problem of autonomous detection of basketball image motion blocks by optical flow method. The optical flow method equation is analyzed, and the change of the motion block pixels is judged by the time domain change and correlation of the intensity data of the pixels in the basketball image sequence. The foreground position of basketball image is extracted by code book model to avoid crowd occlusion interference and find feature points in the foreground position. After obtaining the feature points of the target, the motion blocks are tracked by optical flow method. For all the motion blocks that can be tracked, the optical flow direction data is introduced. The feature points of the optical flow direction in the same angle interval are treated as a group of data to complete normalization, which reduces the misjudgement of the normal motion blocks and enhances the detection precision. Degree. The optical flow acceleration in each interval is processed by Gauss filtering, and the acceleration of each angle interval is accumulated as the basketball image acceleration. The accumulated acceleration threshold is set. When the accumulated acceleration of the image block is higher than the set threshold, the difference is considered. The results show that the proposed method can detect the difference of motion blocks in basketball images under occlusion background, and the values of NMI and LODF are larger. The results show that the proposed method meets the requirements of motion block difference detection under occlusion, and the detection results are accurate, misdetection is not easy to occur.