基于误差补偿的复杂场景下背景建模方法

发布时间：2018-08-20 09:56

【摘要】：在基于子空间学习的背景建模方法中,利用背景信息对前景误差进行补偿有助于建立准确的背景模型.然而,当动态背景(摇曳的树枝、波动的水面等)和复杂前景等干扰因素存在时,补偿过程的准确性和稳定性会受到一定的影响.针对这些问题,本文提出了一种基于误差补偿的增量子空间背景建模方法.该方法可以实现复杂场景下的背景建模.首先,本文在误差补偿的过程中考虑了前景的空间连续性约束,在补偿前景信息的同时减少了动态背景的干扰,提高了背景建模的准确性.其次,本文将误差估计过程归结为一个凸优化问题,并根据不同的应用场合设计了相应的精确求解算法和快速求解方法.再次,本文设计了一种基于Alpha通道的误差补偿策略,提高了算法对复杂前景的抗干扰能力.最后,本文构建了不依赖于子空间模型的背景模板,减少了由前景信息反馈引起的背景更新失效,提高了算法的鲁棒性.多项对比实验表明,本文算法在干扰因素存在的情况下仍然可以实现对背景的准确建模,表现出较强的抗扰性和鲁棒性.
[Abstract]:In the background modeling method based on subspace learning, using background information to compensate the foreground error is helpful to establish an accurate background model. However, when the dynamic background (swaying branches, fluctuating water surface, etc.) and complex foreground factors exist, the accuracy and stability of the compensation process will be affected to some extent. To solve these problems, an incremental subspace background modeling method based on error compensation is proposed. This method can realize the background modeling in complex scene. Firstly, the spatial continuity constraints of the foreground are considered in the process of error compensation, and the disturbance of the dynamic background is reduced while the foreground information is compensated, and the accuracy of the background modeling is improved. Secondly, the process of error estimation is reduced to a convex optimization problem, and the corresponding exact algorithm and fast solving method are designed according to different applications. Thirdly, an error compensation strategy based on Alpha channel is designed to improve the anti-jamming ability of the algorithm to complex foreground. Finally, the background template independent of the subspace model is constructed, which reduces the failure of background update caused by foreground information feedback and improves the robustness of the algorithm. A number of comparative experiments show that the proposed algorithm can achieve accurate modeling of the background in the presence of interference factors and has strong robustness and robustness.
【作者单位】：北京理工大学计算机学院;智能信息技术北京市重点实验室;
【基金】：国家自然科学基金(61273273,61175096,61271374) 高等学校博士学科点专项科研基金(2012110110034) 北京市教委共建项目资助~~
【分类号】：TP391.41

【相似文献】