A Multi-Sensor Acquisition Architecture and Real-Time Reference for Sensor and Fusion Methods Benchmarking

Localization is a key functionality for advance driving assistance systems (ADAS) as well as for vehicle-vehicle or vehicle-infrastructure cooperation. Indeed, depending on the accuracy and integrity of the localization process, applications such as driver information, driver assistance or even fully autonomous driving can be performed. This paper presents a multisensor acquisition architecture for localization. Special attention has been given to main parameters that can affect the accuracy of the localization system. Several sensor technologies have been used and special care to intrinsic, spatial and temporal calibration was given. Since a timestamping synchronizations error induces an error in space on the configuration of a mechanical system, it is necessary to combine synchronized sensor data. A suitable way to handle such problem is to timestamp sensor information in the same time reference frame. The originality of the approach is the use of the SensorHub, a parallel hardware electronic device to perform data acquisition and timestamping in coordinated universal time (UTC). In complement with the SensorHub, the authors demonstrate the real time estimation of a reference trajectory computed from a real time kinematic (RTK) GPS receiver and a hi-grade inertial navigation system (INS) that also timestamp information in UTC time scale. Several sensor databases corresponding to different driving scenarios (environment, speed) were recorded and were used in the future to benchmark a set of fusion methods for localization of road vehicles