2024/12/13 by Zhihang Song, Lihui Peng, Song, Zhihang +7
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Robotics and Sensor-Based Localization #Video Surveillance and Tracking Methods
paper · pdf · doi:10.48550/arxiv.2412.10033
The multi-modal perception methods are thriving in the autonomous driving\nfield due to their better usage of complementary data from different sensors.\nSuch methods depend on calibration and synchronization between sensors to get\naccurate environmental information. There have already been studies about\nspace-alignment robustness in autonomous driving object detection process,\nhowever, the research for time-alignment is relatively few. As in reality\nexperiments, LiDAR point clouds are more challenging for real-time data\ntransfer, our study used historical frames of LiDAR to better align features\nwhen the LiDAR data lags exist. We designed a Timealign module to predict and\ncombine LiDAR features with observation to tackle such time misalignment based\non SOTA GraphBEV framework.\n