Bayesian Gaussian Mixture Models for Enhanced Radar Sensor Modeling: A Data-Driven Approach towards Sensor Simulation for ADAS/AD Development.

In the realm of road safety and the evolution toward automated driving, Advanced Driver Assistance and Automated Driving (ADAS/AD) systems play a pivotal role. As the complexity of these systems grows, comprehensive testing becomes imperative, with virtual test environments becoming crucial, especially for handling diverse and challenging scenarios. Radar sensors are integral to ADAS/AD units and are known for their robust performance even in adverse conditions. However, accurately modeling the radar's perception, particularly the radar cross-section (RCS), proves challenging. This paper adopts a data-driven approach, using Gaussian mixture models (GMMs) to model the radar's perception for various vehicles and aspect angles. A Bayesian variational approach automatically infers model complexity. The model is expanded into a comprehensive radar sensor model based on object lists, incorporating occlusion effects and RCS-based detectability decisions. The model's effectiveness is demonstrated through accurate reproduction of the RCS behavior and scatter point distribution. The full capabilities of the sensor model are demonstrated in different scenarios. The flexible and modular framework has proven apt for modeling specific aspects and allows for an easy model extension. Simultaneously, alongside model extension, more extensive validation is proposed to refine accuracy and broaden the model's applicability.

Development and Experimental Validation of an Intelligent Camera Model for Automated Driving.

The virtual testing and validation of advanced driver assistance system and automated driving (ADAS/AD) functions require efficient and realistic perception sensor models. In particular, the limitations and measurement errors of real perception sensors need to be simulated realistically in order to generate useful sensor data for the ADAS/AD function under test. In this paper, a novel sensor modeling approach for automotive perception sensors is introduced. The novel approach combines kernel density estimation with regression modeling and puts the main focus on the position measurement errors. The modeling approach is designed for any automotive perception sensor that provides position estimations at the object level. To demonstrate and evaluate the new approach, a common state-of-the-art automotive camera (Mobileye 630) was considered. Both sensor measurements (Mobileye position estimations) and ground-truth data (DGPS positions of all attending vehicles) were collected during a large measurement campaign on a Hungarian highway to support the development and experimental validation of the new approach. The quality of the model was tested and compared to reference measurements, leading to a pointwise position error of 9.60% in the lateral and 1.57% in the longitudinal direction. Additionally, the modeling of the natural scattering of the sensor model output was satisfying. In particular, the deviations of the position measurements were well modeled with this approach.

Angle-dependent spectral reflectance material dataset based on 945 nm time-of-flight camera measurements.

The main objective of this article is to provide angle-dependent spectral reflectance measurements of various materials in the near infrared spectrum. In contrast to already existing reflectance libraries, e.g., NASA ECOSTRESS and Aster reflectance libraries, which consider only perpendicular reflectance measurements, the presented dataset includes angular resolution of the material reflectance. To conduct the angle-dependent spectral reflectance material measurements, a new measurement device based on a 945 nm time-of-flight camera is used, which was calibrated using Lambertian targets with defined reflectance values at 10, 50, and 95%. The spectral reflectance material measurements are taken for an angle range of 0° to 80° with 10° incremental steps and stored in table format. The developed dataset is categorized with a novel material classification, divided into four different levels of detail considering material properties and distinguishing predominantly between mutually exclusive material classes (level 1) and material types (level 2). The dataset is published open access on the open repository Zenodo with record number 7467552 and version 1.0.1 [1]. Currently, the dataset contains 283 measurements and is continuously extended in new versions on Zenodo.