Hardware-Triggered Synchronization Method for Robust Camera-Radar Fusion in Autonomous Driving

For robust multi-sensor fusion in autonomous vehicles, where perception algorithms rely on simultaneously collecting data across multiple modalities, precise temporal synchronization of radar and camera sensors is essential. Timestamp-based software synchronization techniques dominate current practice; however, these approaches have poor temporal overlap between sensor acquisition windows and variable delay. This work presents a hardware-triggered synchronization approach that generates Frame Synchronization (FSYNC) signals for camera capture using radar scan timing. An adaptive predictive triggering system was developed using an OnLogic industrial computer with GPIO capabilities, a TIER4 GMSL2 camera, and a Delphi ESR radar. We carried out an experimental validation comparing three methods: the predominantly used software timestamps approach, PTP clock-based synchronization, and hardware triggering. The hardware-triggered approach achieved time gaps of 2.18-4.12 ms across two driving scenarios. In comparison, PTP synchronization showed gaps of 21.17-23.75 ms, while software methods had the largest gaps of 31.353-5.13 ms. Moreover, the hardware approach more than doubled the temporal overlap between radar scan duration and camera exposure window compared to alternative methods, ensuring radar detections and camera pixels capture overlapping time intervals. These results establish hardware synchronization as the optimal method for multi-sensor fusion in autonomous vehicles.