Adaptive multiple video sensors fusion based on decentralized Kalman filter and sensor confidence

Abstract

The fusion of multiple video sensors provides an effective way to improve the robustness and accuracy of video surveillance systems. In this paper, an adaptive fusion method based on a decentralized Kalman filter (DKF) and sensor confidence is presented for the fusion of multiple video sensors. The adaptive scheme is one of the approaches used for preventing the divergence problem of the filter when statistical values of the measurement noises of the system models are not available. By introducing the sensor confidence, we can adaptively adjust the measurement noise covariance matrix of the local DKFs and thus, determine the weight of each sensor more correctly in the fusion procedure. Also, the DKF applied here can make full use of redundant tracking data from multiple video sensors and give more accurate fusion results in an efficient manner. Finally, the fusion result with improved accuracy is obtained. Experimental results show that the proposed adaptive decentralized Kalman filter fusion (ADKFF) method works well in the case of real-world video sequences and exhibits more promising performance than single sensors and comparative fusion methods.

创新点

针对已有的图像融合算法难以在视频传感器有视角、距离、亮度等较大变化的情况下进行融合的问题,本文提出了一种基于分散式卡尔曼滤波的自适应视频传感器融合算法。在融合过程中,该算法采用传感器可信度自动调节局部卡尔曼滤波器的测量误差协方差矩阵,自适应地为待融合传感器分配更加准确的权重。此外,将分散式卡尔曼滤波框架引入到视频传感器融合中,使得来自不同视频传感器冗余的跟踪数据得到充分的利用。ADKFF算法可以有效减少因不正确的目标跟踪和位置映射引起的目标位置误差。