基于雷达与相机融合的行人轨迹跟踪

Pedestrian trajectory prediction by collaborative fusion of radar and camera

针对行人被遮挡时和复杂光线条件下传统跟踪算法易失效的问题,提出一种基于雷达与相机融合的粒子滤波跟踪方法。利用CRFNet神经网络对视觉和雷达底层信号进行融合,基于神经网络对雷达特征的提取,显著提高粒子滤波在夜晚与恶劣气候条件下的跟踪精度。在数据关联过程中,将相机检测到的目标方位角与雷达测量得到的距离相结合,减少传感器误差与模型误差对跟踪的影响。实验结果表明,该算法相比传统融合跟踪算法在精度上提高了10%,鲁棒性显著增强。

The traditional tracking algorithm is easy to tail when pedestrians are occluded or under complex light conditions.Targeting at these problems,a particle filter tracking method based on the fusion of radar and camera sensors was put forward.By taking advantage of the CRFNet neural network,the underlying signals of vision and radar sensors were fused and the radar features were extracted to significantly improve the tracking accuracy of the particle filter at night and under bad weather conditions.In the data correlation process,the combination of camera-detected target azimuth and radar-measured distance greatly reduced the impact of sensor error and model error on tracking.According to the experimental results,the proposed algorithm improves the tracking accuracy by 10%together with the remarkable robustness enhancement compared to the traditional fusion tracking algorithm.