In August, 2021 our team has performed various measurements in relation to environment perception and understanding at ZalaZONE proving ground (located near the city of Zalaegerszeg in Hungary) by relying on a distributed sensory network (composed of sensors of various types) deployed by our team.
The primary aim of the tests was to evaluate how the certain sensors perceive the environment (including static and dynamic objects) under various conditions (such as partial overlapping, various distances, variable visibility conditions, sensors deployed at different heights, various formations of vehicle groups, etc.). On top of that, an additional goal was to evaluate how the fusion based detection and tracking algorithms - developed by our team – perform on such collected raw sensory data.
Regarding the above-mentioned distributed sensory network, we have deployed two sensor stations equipped with different type of LiDARs, RADAR, camera, thermal vision camera and a GPS time source used for synchronization purposes. The setup can be followed in Fig.1. As one can see the test region is a motorway section, which is located at the ZalaZONE proving ground.
One of the stations was deployed on a bridge above the motorway at a height of about 5m while the other one on the motorway itself in order to test how the mounting height affects the performance of sensors and that of the detection algorithms. The extrinsics of all sensors have been estimated with respect to the UTM frame.
Regarding the algorithms to be evaluated, we have focused on raw fusion based vehicle detection algorithms - developed by our team - but also on the detector used by the RADAR itself (see Fig. 2). In order to evaluate the accuracy and reliability of detections, static as well as dynamic reference data are highly welcome.
During the tests all the participating vehicles were equipped with a high precision differential GPS system in order to offer reference data (position, heading, timestamp) for evaluation of the perception algorithms (see Fig. 3). The collected data cover various scenarios executed with several vehicles during daytime as well as in the night. In addition to such generated dynamic reference data, static ones in form of a dense HD map of the motorway section is also available. This offers an elegant way for evaluating not only the sensors but also the performance of perception algorithms under various conditions, which may significantly contribute to their further improvement.