Integration of a lane detection system in a virtual environment to test and evaluate active safety and autonomous driving - A research project at Volvo cars to emulate the behavior of a lane detection system

Abstract

A vehicle's ability to determine its position relative to the road is a crucial feature for active safety functions and autonomous driving. Consequently, autonomous vehicles and current vehicles featuring active safety systems are usually equipped with a lane detection system that estimates the position of the vehicle relative to the road lane markings. The field of active safety has grown exponentially during the past decade, which in turn has hugely increased the demand for functional testing. Virtual environments are particularly economical, rapid, and safe for functional testing and are today also used to simulate autonomous driving. The aim of this thesis was to design and implement a model for a visual-based lane detection system to be integrated in the virtual environment used for functional tests at Volvo Cars. A virtual camera model with appurtenant camera algorithms was developed and enabled the computation of a 2D-image through projection of points in 3D-space from the virtual environment, mimicking the behavior of a real visual lane detection system. To emulate the real performance of the sensor, various software components were mod- eled and their outputs compared to field data from a production lane detection system. A multivariate analysis was executed to characterize the real system's error and improve the virtual model. The error performance was modeled with a 3-dimensional confidence curve together with a low pass FIR filter to create the extreme boundaries of the sensor. The model of the system behavior was merged with the camera model to create a vision based virtual lane detection system. This virtual system was integrated and verified in the virtual environment for functional test at Volvo Cars.