基于压缩感知的多目标实时跟踪系统

发布时间：2018-07-06 14:03

本文选题：压缩感知 + 目标检测　；参考：《北京邮电大学》2016年硕士论文

【摘要】：多目标实时跟踪是计算机视觉领域的研究热点之一,在智能交通、智能监控等多个领域有着广泛的应用,然而设计一个鲁棒性高的多目标实时跟踪系统,无论是对科学研究还是工程实践都极具挑战性。压缩感知算法通过对信号采样压缩,能够大大降低信号的复杂度。将压缩感知理论与多目标实时跟踪相结合能够提升系统跟踪的稳定性和实时性。本文基于压缩感知理论,设计了一套鲁棒性高的多目标实时跟踪系统。主要研究内容如下:(1)研究并实现了基于Haar特征的AdaBoost目标检测器。提取视频图像的Haar特征,并基于大量多角度人头正负样本图像训练AdaBoost级联分类器,实现多目标检测。(2)设计并改进了基于压缩感知的朴素贝叶斯目标跟踪器。基于压缩感知理论,对目标特征采样获得压缩感知特征,并构建分特.征权重的朴素贝叶斯在线学习分类器,实现多目标实时跟踪。(3)开发了一套多目标实时跟踪系统,在多种实验场景下与MHT算和CT算法进行检测跟踪实验比较,证明本系统在实时性、准确性和稳定性上都有较好的表现。本文将压缩感知算法应用到目标跟踪系统中,实现了目标跟踪系统稳定性和实时性的兼备的多目标实时跟踪系统,具有很高的工程研究价值和社会应用价值。
[Abstract]:Multi-target real-time tracking is one of the hotspots in the field of computer vision. It has been widely used in many fields, such as intelligent transportation, intelligent monitoring and so on. However, a robust multi-target real-time tracking system is designed. Both scientific research and engineering practice are extremely challenging. The compression sensing algorithm can greatly reduce the complexity of the signal by sampling and compressing the signal. Combining compression sensing theory with multi-target real-time tracking can improve the stability and real-time of system tracking. Based on the theory of compressed sensing, a robust multi-target real-time tracking system is designed in this paper. The main contents are as follows: (1) AdaBoost target detector based on Haar feature is studied and implemented. The Haar feature of video image is extracted and the AdaBoost cascade classifier is trained based on a large number of multi-angle head positive and negative samples. (2) A naive Bayesian target tracker based on compressed sensing is designed and improved. Based on the theory of compressed perception, the compressed perceptual features are obtained by sampling the target features, and the sub-features are constructed. Naive Bayesian online learning classifier with eigenweight is used to realize multi-target real-time tracking. (3) A multi-target real-time tracking system is developed and compared with MHT algorithm and CT algorithm in various experimental scenarios. It is proved that the system has good performance in real time, accuracy and stability. In this paper, the compressed sensing algorithm is applied to the target tracking system, and the multi-target real-time tracking system is realized, which is both stable and real-time. It has high engineering research value and social application value.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】