多传感器航迹融合能够在一定准则下整合数据，获得比单一的传感器数据更接近被测物理量真值的状态估计值和更有价值的综合信息。目前，卡尔曼滤波算法和传统的加权平均融合算法虽然能有效地实现航迹融合，但融合所得测量精度仍然有限。文中针对航迹数据特点，提出基于多输出的多传感器在线航迹融合算法，通过多输出多元线性回归模型实现了航迹数据融合；利用云边协同架构完成模型训练和推理，在中心云端利用历史数据训练得到多输出多元线性回归模型，在此基础上，在边缘端利用在线梯度下降算法完成模型权重系数的实时更新。实验结果表明，文中提出的算法在融合精度上要优于目前的卡尔曼滤波算法和传统的加权平均融合算法。

Multi-sensor track fusion can integrate data under certain criteria, and obtain state estimation values and more valuable comprehensive information that are closer to the true values of the measured physical quantities than a single sensor data. At present, although the Kalman filter algorithm and the traditional weighted average fusion algorithm can effectively realize track fusion, the measurement accuracy obtained by fusion is still limited. According to the characteristics of track data, a multi-sensor online track fusion algorithm based on multi-output is proposed in this paper, and the multi-output multiple linear regression model is used to realize the track data fusion. The cloud-edge collaborative architecture is used to complete the model training and inference, and the multi-output multiple linear regression model is trained by using historical data in the central cloud, moreover the online gradient descent algorithm is used at the edge to complete the real-time update of the model weight coefficient on this basis. Experimental results show that the proposed algorithm is better than the current Kalman filter algorithm and the traditional weighted average fusion algorithm in terms of fusion accuracy.

TP181