视觉量子目标跟踪方法

为了解决变结构目标跟踪过程中目标失跟率较高的问题，提出了一种基于视觉量子（vision quantum，简称VQ）的目标跟踪方法.该方法首先在图像内自上而下地辐射视觉量子采集灰度信息，统计量子内部概率密度较大的灰度级和分布区域；然后计算视觉量子的量子频率，归一化量子频率系数，滤除系统噪音和杂波干扰，利用频率阶跃不变性移动视觉量子至平衡状态，将达到量子平衡状态的视觉量子组成量子簇；最后，以该量子簇作为候选目标信息，采用极大似然估计预测运动目标状态，以预测结果作为下一帧图像中视觉量子移动的参考值，并进一步验证移动后的视觉量子是否达到量子平衡状态，以确保目标跟踪有效性.该方法抓住了变结构运动目标前景与背景交界处具有量子频率阶跃不变性的特点，继而将阶跃不变特征采用具有独立性和约束性的视觉量子进行描述，可以有效地消除形状变化、尺度变化等变结构因素对运动目标跟踪的影响，失跟率较低.同时，由于视觉量子数据量较小，计算复杂度较低，其跟踪实时性较高.大量实验测试结果表明，该方法对变结构目标跟踪具有很好的适应性、实时性和鲁棒性.

Object Tracking Method Based on Vision Quantum

An approach to object tracking based on vision quantum is proposed in this paper in order to solve the high loss-tracking rate in variable structure object tracking. First, the gray information is detected in an image from top to bottom with vision quantum, and the distribution area and gray levels of larger probability density are counted in the vision quantum. Then all the energy frequencies of the visual quantum are calculated such that the weaker energy frequency gradient is removed by filtration and the stronger frequency gradient of vision quantum that the distribution of high frequency information account for half quantum area is reserved. The quantum cluster is composed of vision quantum with the same frequency variation. Finally, taking quantum cluster as candidate object information, the state of moving object is predicted with maximum likelihood estimation and the forecast results are served as moving reference position of vision quantum in the next frame. Further verification of the visual quantum balance state is made to ensure the effectiveness of object tracking. This method catches the point that the variable structure moving object has the feature of the energy frequency step invariance at the juncture pixels of the foreground and background. It can effectively overcome the changes in shape, scale and other factors that influence the moving object tracking, achieving lower loss-tracking rate and lower computational complexity by using independent and continuous visual quantum to describe the step invariant feature. Experimental results show that the proposed approach has good adaptability to variable structure tracking with real-time and robust tracking performance.