基于遮挡检测的粒子滤波行人目标跟踪算法研究

发布时间：2018-11-06 14:15

【摘要】：在过去的几十年中,行人目标跟踪技术已经成为了计算机视觉领域最热门的课题之一。许多学者对行人目标跟踪技术进行了大量的研究,促进了其的快速发展。但是当把视频序列应用到现实生活中时,跟踪器需要处理很多难题,例如遮挡、相似物干扰、光照变化、尺度变化以及背景杂乱等。在这些难题中,遮挡是最棘手的问题之一。基于此,为解决复杂遮挡情况下的行人目标跟踪问题,本文对行人目标跟踪中的外观模型建模、状态估计和遮挡检测等几个关键技术问题进行了深入研究。具体研究内容如下:针对行人目标在遮挡状态下容易产生跟踪漂移的问题,提出一种基于重构误差方差的粒子滤波行人目标跟踪算法(VREPT)。该算法在粒子滤波跟踪框架下,采用主成分分析方法建立目标外观模型;然后,为了处理遮挡干扰问题,在跟踪系统中引入遮挡检测器,对视频序列中每一帧的跟踪结果,根据重构误差的方差检测目标是否处于遮挡状态,并依据检测结果进行相应的遮挡处理;当检测到目标处于被遮挡状态时,目标模板将停止更新,当目标未被遮挡时,采用一种增量方式有效学习和更新目标外观模型。实验结果表明,提出的算法能够有效地处理遮挡,并且能够对目标进行准确稳定地跟踪。针对行人目标跟踪中目标遮挡的快速检测问题,提出一种基于高阶累积量的行人目标跟踪算法(HOCPT)。在提出的算法中,利用高阶累积量能够有效抑制高斯噪声的特性,构造重构误差的三阶累积量,并以此构建目标遮挡检测器对行人目标进行遮挡检测,实现对遮挡目标的快速实时检测和处理。实验结果表明,提出算法能够准确而快速地检测到目标进入和离开遮挡的时刻,并对遮挡问题进行有效的处理,具有稳健的跟踪效果。
[Abstract]:In the past few decades, pedestrian target tracking technology has become one of the hottest topics in the field of computer vision. Many scholars have done a lot of research on pedestrian target tracking technology to promote its rapid development. However, when video sequences are applied to real life, the tracker needs to deal with many problems, such as occlusion, similarity interference, illumination change, scale change and background clutter. Of these problems, occlusion is one of the thorniest. Based on this, in order to solve the problem of pedestrian target tracking in the case of complex occlusion, several key technical problems such as appearance model modeling, state estimation and occlusion detection in pedestrian target tracking are studied in this paper. The specific research contents are as follows: aiming at the problem that pedestrian target is easy to track drift in occlusion state, a particle filter pedestrian target tracking algorithm (VREPT). Based on reconstruction error variance is proposed. In the framework of particle filter tracking, the method of principal component analysis (PCA) is used to establish the object appearance model. Then, in order to deal with the occlusion interference problem, the occlusion detector is introduced into the tracking system. The tracking results of each frame in the video sequence are detected according to the variance of the reconstruction error. According to the test results, the corresponding occlusion treatment is carried out. When the target is detected to be occluded, the target template will stop updating, and when the target is not occluded, an incremental approach will be used to effectively learn and update the target appearance model. Experimental results show that the proposed algorithm can deal with occlusion effectively and can track the target accurately and stably. Aiming at the problem of fast detection of target occlusion in pedestrian target tracking, a pedestrian target tracking algorithm (HOCPT). Based on high order cumulants is proposed. In the proposed algorithm, the feature of Gao Si noise can be effectively suppressed by using high-order cumulants, and the third-order cumulant of reconstruction error can be constructed, and the target occlusion detector is constructed to detect pedestrian targets. The fast real-time detection and processing of occlusion targets are realized. Experimental results show that the proposed algorithm can accurately and quickly detect the entry and departure of the target occlusion, and effectively deal with the occlusion problem, and has a robust tracking effect.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】