A Hybrid Timestamping Approach for Multi-Sensor Perception Systems

Abstract

Synchronized and precisely timestamped data from perception sensors is highly advantageous for the low-level fusion of multiple sensor data. Many open-available, low-cost perception sensors do neither provide hardware support for precise clock synchronization, nor provide timestamps with their measurement data. In this work, we present an approach to enable synchronization and accurate timestamping of hardware-triggerable sensors in multi-sensor perception systems.We utilize a hybrid timestamping approach, taking into account the timestamp of a hardware trigger and the software timestamp. The presented timestamping approach utilizes the trigger time to assign precise timestamps to the data streams of the perception sensors. Precise timestamps are mandatory in order to achieve a high perception performance in dynamic applications which utilize low-level data streams.Additionally, we present an implementation of the approach on a multi-sensor perception platform, archiving a timestamp precision in the range of 2 ms. An existing Robot Operating System (ROS) architecture of the platform is extended to assign hybrid timestamps to the data streams. Additionally, we present a pedestrian detection implementation which fuses the timestamped data into a representation.