基于支持向量机的多传感器数据融合算法

摘要

提出一种基于支持向量机(SVM)的多传感器数据融合算法提高融合性能，分别用3种核函数作为特征空间的内积。结合扩展卡尔曼滤波器(EKF)和软计算原理，在雷达/红外多传感器跟踪系统中构建一个有效的信息融合框架。引入环境信息使传感器信任度预测适应环境的变化，降低多传感器系统不确定因素的影响；联合测量方差归一化变量(NVMSE)作为支持向量机的输入，训练得到高确定性和高精确度的传感器信任度预测。仿真结果表明，与传统的多传感器数据融合算法相比，该算法性能更佳。%An efficient multi-sensor data fusion framework based on support vector machine (SMV)was proposed,three kernel functions were used for comparison.Expanded Kalman filter (EKF)and soft computing principle were combined to structure a data fusion method with high performance for radar/infrared sensor tracking system.Context information was utilized to make the estimated sensor confidence degree more adaptive to contextual changes,reducing the uncertain and unsettling influence of the multisensory data estimation.Normalized variables of the measurement square error was unified as the input,it performed an improvement in predictive accuracy and generalization capacity.Results of simulation show the proposed method is better than classic fusion algorithms.