针对视频图像中多目标运动、边缘特征模糊、目标跟踪难度大的问题,提出了一种基于多目标视频图像边缘特征的核相关滤波跟踪算法。首先,将视频图像中目标运动轨迹的3帧图像时间作为线性段。然后,利用线性判断方法捕获目标,利用动态边缘演化技术准确提取捕获目标的边缘特征;并结合视频图像梯度角度直方图与颜色信息,获取梯度角度-色度饱和度直方图颜色特征,得到跟踪目标的特征权重。最后,利用核相关滤波跟踪算法,通过循环移位和循环矩阵、岭回归模型学习分类器实现视频图像的多目标跟踪。实验结果表明,本算法的多目标跟踪成功率高达99%以上,且在尺寸变化、颜色变化、存在遮挡物等复杂环境下每秒能跟踪的图像数量大于65 frame,具有优越的跟踪性能。

Considering the problems of moving multi-target in video images, fuzzy edge features, and difficult target tracking, a kernel correlation filtering tracking algorithm based on edge features of multi-target video images is proposed in this paper. First, the time of 3 frame images of the target motion trajectory in video images is set as the linear segment. Then, the linear judgment method is used to capture the target. In addition, the dynamic edge evolution technology is used to accurately extract the edge features of the captured target; combined with the gradient angle histogram and color information of video images, the gradient angle-chroma saturation histogram color features are obtained, and the feature weight of the tracking target is obtained. Finally, the kernel correlation filtering tracking algorithm is used to realize the multi-target tracking of video images through cyclic shift, cyclic matrix, and ridge regression model-learning classifier. The experiment results show that the multi-target tracking success rate of the algorithm is above 99%, and the number of images that can be tracked per second is above 65 frames in the complex environment, such as size change, color change, and occlusion, which has superior tracking performance.