压缩感知跟踪中的特征选择与目标模型更新

Feature selection and target model updating in compressive tracking

目的为了增强压缩感知跟踪算法在复杂场景下的性能,提出一种特征选择与目标模型更新的改进跟踪算法。方法本文算法包含两方面的改进,一是根据特征的正负类条件概率分布的距离选择能有效区分目标与背景的特征;二是根据当前目标与原始目标的差异自适应更新目标外观模型,使得目标遇到较大遮挡或者姿态频繁改变时目标外观模型不会被错误更新。结果对于10个复杂环境下的经典视频序列,基于压缩感知的改进跟踪算法获得平均85.19%的正确跟踪率和平均13.74个像素的跟踪误差效果,在中心误差、成功率和精确度3个指标上均优于最近提出的3个代表性跟踪算法。结论实验结果表明,本文新的特征选择和目标模型更新算法,既增强了压缩感知跟踪算法的鲁棒性,又加快了跟踪速度。

Objective In order to enhance the performance of compressive sensing based tracking in complex scenarios, we propose an improved tracking algorithm based on a new feature selection approach and target model updating mechanism. Method First, we select features allowing to distinguish a target from the background, according to the distance between a feature＇s positive and negative class conditional probability Gaussian distributions. Second, we adaptively update the target appearance model according to the difference between the current target and the original one, so that the target would not be updated in case of big occlusion or frequent posture changes. Result Experiments on ten standard and complex test video sequences demonstrated that for the three measurements, i.e. center error, the success rate, and the precision plot, our algorithm, with the rate of 85.19% of frames correctly tracked and average 13.74 pixels of center location difference, achieves a higher perform than three state-of-the-art methods. Conclusion The proposed new method of feature selection and target model updating, enhances the robustness of compressive sensing based tracking and speed up of the track.