相关滤波跟踪算法的改进预处理技术

Improved Preprocessing Technology for Tracking Algorithm based on Correlation Filtering

随着人工智能科学技术的发展,计算机视觉目标跟踪愈来愈受到广泛关注,在生活上的应用方面也逐渐铺开,如人机交互、无人驾驶等。近年来,视觉目标跟踪的发展取得了重大进步,但是仍存在许多问题需要进一步完善。基于相关滤波技术的跟踪方法通常使用基于高斯分布的余弦窗口对目标进行预处理。这种方法的缺点在于当目标的状态变化较大时,算法以上一帧目标位置对目标进行采样并预处理,此时算法会虚弱目标信息并增强目标背景信息。因此本文提出一种改进的基于贝叶斯的概率图谱预处理技术来解决这个问题,以实现更加准确地反映目标的前景信息。

With the development of artificial intelligence science and technology, vision object tracking is more and more widely concerned and its application in life is also gradually spread, such as human-computer interaction, unmanned driving, and so on. Recently, the development of visual target tracking has made great progress. But there are still many problems to be improved. For example, a cosine window based on Gaussian distribution is usually used to preprocess the target in the tracking algorithm based on correlation filtering technology. Its disadvantage is when the state of the target changes greatly, the algorithm samples and preprocesses the target according to its position in the previous frame. So the target information is weakened and its background information is enhanced. Therefore, this paper proposes an improved preprocessing technology to solve this problem using a probability map based on Bayesian, which can reflect the foreground information of the target more accurately.