为了解决目标跟踪中运动模型复杂,运动场景多变的情况,提出了一种多模式融合的目标跟踪算法。该算法选取目前广泛应用的mean-shift和粒子滤波算法分别跟踪目标,得到当前目标位置的候选值,并采用加权合成参考函数建立参考模板。然后,以侯选目标位置差异和参考模板为标准,确定目标的正确位置。最后,根据当前帧目标模板和参考模板的距离来决定是否更新模板。实验仿真结果表明,与单一的目标跟踪算法相比,本文算法的平均跟踪误差减小了一倍以上。假如参考模板更新错误,下一帧中仍能以67%的概率正确跟踪目标,连续3次模板更新之后,误更新的模板对目标跟踪的影响可以降低到10%以下,有效地降低了模板更新引起的跟踪错误和跟踪不稳定。

In accordance with complicated object movement and changeable object environment, a multi-pattern fusion algorithm for target tracking is presented. Mean-shift and particle filter algorithms widely applied to target tracking are selected to get tentative locations and Weighted Composite Reference Function (WCRF) is adopted to establish reference model. Then, the distance difference of the tentative locations and the reference model is considered as a criterion to find correct location. Finally, the algorithm updates the reference model according to the distance between reference model and target model in current frame. The experimental simulation results show that the average tracking error of the proposed algorithm is reduced by 50% as compared with that of single target tracking method. If the reference model is updated incorrectly, the probability to find the correct location in the next frame is 67%. After updating the reference model three times, the influence on object tracking is less than 10%,which effectively reduces the tracking error and instability for model updating.