基于卡尔曼滤波的动态轨迹预测算法

A Dynamic Trajectory Prediction Algorithm Based on Kalman Filter

基于拟合的传统轨迹预测算法已无法满足高精度和实时性预测要求.提出基于卡尔曼滤波的动态轨迹预测算法,对移动对象动态行为进行状态估计,利用前一时刻的估计值和当前时刻的观测值更新对状态变量的估计,进而对下一时刻的轨迹位置预测.大量真实移动对象数据集上的实验结果表明:Geo Life数据集上基于卡尔曼滤波的轨迹预测算法的平均预测误差(预测轨迹点与实际轨迹点的均方根误差)为12.5米;与基于轨迹拟合的轨迹预测算法相比,T-Drive数据集预测误差平均下降了555.4米,预测准确率提升了7.1%.在保证预测时效性前提下,基于卡尔曼滤波的动态轨迹预测算法解决了轨迹预测精度较低的问题.

Traditional fitting-based trajectory prediction algorithms cannot meet the requirements of high accuracy and real-time prediction. A dynamic Kalman filter based TP approach was proposed, which performs state estimation of dynamic behavior with regard to moving objects, updates the state variable estimation value based on the estimation of the previous and current observation states, in order to infer the next location of moving objects. Extensive experiments are conducted on real datasets of moving objects and the results demonstrate that the average prediction error （ root mean square error between the predicted location and the actual location） of the TP algorithm based on Kalman filter is around 12.5 meters on the Ge- oLife datasets. The prediction error is reduced by about 555.4 meters by compared to the fitting-based TP algorithms, and the prediction accuracy is increased by 7.1% on the T-Drive datasets as well. The dynamic TP approach based on Kalman filter can handle the problem of low prediction accuracy with the guarantee of efficient time performance.