Disturbance-resistant Camera-LiDAR Fusion for Robust Three-dimentional Object Detection.

Three-dimentional object detection using monocular cameras and LiDAR fusion presupposes that the accuracy of extrinsic calibration between cameras and LiDAR is satisfactory. However, in engineering, the Camera-LiDAR calibration error is always quite large. In this paper, the fusion problem of LiDAR-camera with perturbed calibration parameters is investigated, and a soft-association method is proposed for heterogeneous sensor data association. Experiments with the KITTI datasets are carried out to verify our proposed method. Under 6 degrees of freedom bias perturbations, the intersection-over-union score obtained from our method is higher than that of the traditional method. In addition, our method allows valid correlations with translation errors of $\pm 15$ mm and rotation errors of $\pm 0.06$ rad, and our algorithm is successfully deployed on low resolution LiDAR and camera platforms.