Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract. Highly automated machines with sensor-based assistance systems can increase performance and efficiency while reducing environmental impact. Especially in agriculture, the environment is characterized by harsh and changing conditions such as weather and vegetation. However, for the safe operation of automated and autonomous machinery, robust environmental perception that works reliably in all environmental conditions is essential. Domain-specific test data is needed to validate the robustness of sensor systems that detect obstacles, such as humans, in an agricultural context.

This paper presents a test environment to investigate the influence of different environmental conditions on the reliability of sensor systems used for semantic environment perception. Sensors, like cameras and LiDAR, are reproducibly moved on a 100-meter-long L-shaped rail-based carrier system. On three different cultivable field areas, e.g. row crop, grassland and fallow, different scenarios with a detection range of up to 30 meters are covered. With automated test runs, domain specific datasets with multiple environmental conditions can be generated to systematically evaluate the accuracy of AI algorithms for environment perception. Therefor the ground truth is automatically generated based on two surveyed pedestrian test targets and the localization of the carrier system. The test field described here is one of three test levels that forms a basis for the development of certification processes together with separate individual tests and a transfer to real agricultural machinery.