基于高维SIFT改进隐马尔可夫模型的多目标跟踪

Multi-object tracking based on improved hidden Markov model with high dimensional SIFT

针对多目标跟踪领域中出现的遮挡、目标身份互换等问题,本文提出了一种基于尺度不变特征变换(Scale-Invariant Feature Transform,SIFT)关键点和隐马尔可夫模型的多目标跟踪算法。首先,在视频序列中逐帧提取每个目标的关键点集,并对其进行条件约束;其次,以得到的关键点集作为目标的状态建立隐马尔可夫模型,根据模型在时序段内传递状态的规律求出模型对应的参数;最后,以当前帧的观测状态和参数求出下一帧的隐性状态,实现对目标位置的预测。为了提升模型的推理速度,建立了表征全部目标的高维观测状态模型。与其他先进的算法在MOT17、MOT20、KITTI数据集上进行了仿真实验对比,结果表明本算法在跟踪准确度等指标上表现较优,并对遮挡和身份互换问题有较好的鲁棒性。

Aiming at the problems of occlusion and object identity exchange in the multi-object tracking,this paper proposes a multi-object tracking algorithm based on hidden Markov model with Scale-Invariant Feature Transform(SIFT)features.Firstly,the SIFT features set of each object in each frame are extracted and constrained.Then the constrained SIFT set is taken as the object state,and a hidden Markov model is established with the state.The corresponding parameters of the model are obtained according to the law of the state transmitting in each period.Finally,the hidden state in the next frame is estimated by the observation state and parameters in the current frame so as to realize the prediction of object location.In addition,in order to improve the reasoning speed of the model,a high-dimensional observation state model representing all objects is established.Compared with other algorithms on MOT17、MOT20 and KITTI datasets,the experimental results show that this algorithm performs better in tracking accuracy and other indicators,and has good robustness to occlusion and identity exchange problems.