Procedural Generation of Vegetation for a Virtual Test Track

Abstract

Advanced driver assistant systems (ADAS) are increasingly being tested during simulated test drives in a test and training environment based on a driving simulator, in order to reduce the number of extensive real test drives. The need for numerous virtual test drives in the driving simulator requires to model detailed and realistically appearing 3D models of real test tracks. A manual reproduction of real tracks is a cumbersome and time-intensive task. In previous work, we have introduced a method to create virtual test tracks with minimized manual effort using data from various sources, such as navigation systems, digital elevation models, aerial images, digital landscape models etc. [1]. However, these virtual test tracks still do not appear very realistic to the test driver, since no detailed vegetation was generated by that method. In this paper, we propose an approach to enrich a virtual terrain with authentic vegetation. The aim is to increase the perceived realism of the landscape, in order to provide the same input for the sensors of an ADAS under test in the driving simulator as on the real track. The requirement is to automate the vegetation generation as far as possible and to support real-time rendering of the generated very complex 3D model, which is crucial for a usable sensor feed. The basis for the generation of vegetation in this work is data from digital landscape models. These data define where areas like woodlands and agricultural zones are located in geographic coordinates. These areas are refined by a color detection, which is applied to the corresponding aerial images, in order to identify various tree and plant species. Based on the application of a procedural rule system the actual plants are then placed in the refined areas. The rule system imitates the natural growth behavior of plants and is based on terrain characteristics like gradient, direction of a slope, or competition for resources. By combining terrain data, color detection on aerial images, and procedural rules, a planting method is developed to generate natural looking vegetation. The implementation prototype of our approach, based on the Unity3D game engine, which supports an easy creation of complex sceneries, showed that it is possible to create vegetation for a virtual test track with minimal manual effort. By placing vegetation at realistic locations, considering natural spread of plants, the perceived realism of the scene was improved. A performance analysis showed that even with the generated vegetation, interactive frame rates are achievable.