基于多特征与尺度自适应的目标跟踪算法

Target Tracking Algorithm Based on Muti-featrue and Scale Adaptation

针对实际应用场景中跟踪目标发生尺度变化、光照变化、背景复杂等问题,提出了基于多特征与尺度自适应的目标跟踪算法。针对不同特征在不同环境下的优势,将颜色直方图特征和深度特征进行了融合;针对目标跟踪过程中出现的尺度变化,引入了尺度滤波器对目标尺度进行判断,选择最合适的目标尺度;同时使用上下文感知滤波器作为框架,提高了上述算法对目标和背景的判别能力;提出了一种新的模型更新判别依据APCE,只有当APCE值和响应最大值同时分别以一定比例大于各自的历史平均值时,才对模型进行更新;最后将所提算法与近年来一些性能优异的目标跟踪算法进行实验对比。实验结果表明,所提算法能够有效的提高该算法在尺度变化、光照变化、背景复杂等环境下的跟踪性能,整体性能优于对比算法。

Aiming at the problems of scale changes,illumination changes,and complex backgrounds of tracking targets in practical application scenarios,a target tracking algorithm based on multiple features and scale adaptation is proposed.In view of the advantages of different features in different environments,the color histogram feature and the depth feature were fused;For the scale changes that occur in the target tracking process,a scale filter was introduced to judge the target scale and select the most suitable target scale;At the same time,the context-aware filter was used as a framework to improve the algorithm’s ability to discriminate between targets and backgrounds;A new model update discrimination basis APCE was proposed,only when the APCE value and the maximum response value were both larger than their respective history by a certain proportion.The model was updated only when the average valuewas reached.Finally,the algorithm in this paper was compared with some excellent target tracking algorithms in recent years.The experimental results show that the proposed algorithm can effectively improve the algorithm in scale changes,lighting changes,and complex backgrounds.The overall performance of tracking performance in other environments is better than the comparison algorithm.