论文部分内容阅读
针对传统交互式多模粒子滤波(IMMPF)跟踪器概率计算方法在复杂环境下鲁棒性不足的缺陷,提出一种新型的基于联合似然函数模型的子跟踪器概率计算方法。首先,计算基于跟踪结果与当前外观模型的巴氏距离作为瞬时似然函数,度量目标外观的剧烈变化;其次,利用l2范数规则化最小二乘算法构建目标的重构外观模型,将其与跟踪结果的误差指数函数作为平稳似然函数,度量目标的缓慢变化;然后,基于加权求和策略得到跟踪器基于多种特征的联合似然函数;最后,将建立的联合似然函数结合上一帧的先验状态交互概率完成子跟踪器概率的更新。对复杂环境下跟踪器性能的在线评估对比结果验证了联合似然函数模型能有效评估跟踪器因不同干扰因素导致的性能变化,将其应用于子跟踪器概率的计算能获得比主流算法更好的鲁棒性。
Aiming at the shortcomings of traditional IMMPF tracker’s robustness in complex environment, this paper proposes a new method to calculate the sub-tracker probability based on the joint likelihood function model. Firstly, we calculate the Pahscholars’ distance between the tracking result and the current appearance model as the instantaneous likelihood function to measure the dramatic change of the appearance of the target. Secondly, construct the reconstructed appearance model of the target by using the l2 norm regularization least squares algorithm, The error exponential function of the tracking result is used as a stationary likelihood function to measure the slow change of the target; then, the joint likelihood function of the tracker based on the multiple features is obtained based on the weighted summation strategy; finally, the combined likelihood function is combined with the previous one The prior state transition probability of the frame completes the update of the sub-tracker probability. The results of online evaluation of tracer performance in complex environment verify that the joint likelihood function model can effectively evaluate the performance changes of trackers due to different interference factors, and the calculation of the tracer probability can be better than the mainstream algorithm Robustness.